From Zero to Hero: Getting Started with GraphQL, Strapi and Next.js 16 Part 4

Part 4 of a 4-part series on building with GraphQL, Strapi v5, and Next.js 16. Each part builds directly on the project from the previous post.

• Part 1, GraphQL basics with Strapi v5. Fresh install, a full Shadow CRUD tour, and your first custom resolvers.
• Part 2, Advanced backend customization. A Note + Tag model, middlewares and policies, custom queries, and custom mutations.
• Part 3, Next.js 16 frontend. Apollo Client on the App Router: Server Component reads, Server Action writes, one page per GraphQL operation.
• Part 4 (this post), Users and per-user content. Sign-in, automatic ownership on every note, an is-note-owner policy for GraphQL writes, the gap that policy leaves on REST, and a Document Service middleware that closes both APIs from one file.

After Part 3 you have a note-taking app that anyone can use. In Part 4 every user signs in, sees only their own notes, and cannot read or change anyone else's, no matter which API they call.

TL;DR

• Strapi's users-permissions plugin gives us JWT sign-in, a User content type, and the register, login, and me GraphQL operations. We turn it on and grant the right permissions. We do not write auth code ourselves.
• Notes get an owner relation to User. A Document Service middleware in src/index.ts register() writes owner automatically on every new note, taking the value from ctx.state.user (the signed-in user). The client cannot pass someone else's id; the middleware overwrites the field at the data layer. The same code path covers REST, GraphQL, custom resolvers, and the seed script.
• Reads work the same way. A second clause in the same middleware adds owner = ctx.state.user.id to the filter of every findMany and findOne on Note. REST GET /api/notes and GraphQL Query.notes both go through the Document Service, so one rule limits each user to their own notes on both APIs.
• Writes get two layers of protection. The read filter above means a non-owner cannot even load the target note, so any update or delete attempt against someone else's note misses. On top of that, an is-note-owner policy on the four GraphQL write mutations gives us a worked policy example and a second check before the resolver runs.
• The custom queries from Part 2 (searchNotes, noteStats, notesByTag) build their own filter object or query a different content type, so the Document Service middleware does not cover them. Each one gets an inline owner check. Part 4's Step 6.3 covers that.
• The frontend gets three small things: /login and /register pages backed by Server Actions that set an HTTP-only cookie, an Apollo link that adds Authorization: Bearer <jwt> to every request, and a Next.js middleware.ts that sends signed-out users to /login.

Backend authorization at a glance:

REST and GraphQL both end up calling strapi.documents("api::note.note").<action>(...), so a rule placed there fires for both APIs.

Prerequisites

• The Part 2 backend running on http://localhost:1337/graphql, with the Note + Tag schema, the soft-delete middlewares, the cap-page-size policy, and the test script (scripts/test-graphql.mjs) all green.
• The Part 3 frontend cloned and working against that backend, including the seeded notes and the home page rendering them. If you skipped Part 3, clone the starter template and run through Steps 1 through 5.
• Node.js 20.9+. Strapi 5 and Next.js 16 both require it.

Scope

Part 4 covers backend authorization end-to-end and the minimum frontend wiring to test it. Skipped:

• Social login (OAuth providers). Email + password only.
• Password reset and email verification. Not part of the authorization story. See the users-permissions docs when you need them.
• Roles beyond owner / not-owner. No admin, no read-only viewer, no sharing. The pattern below is the foundation; richer roles use the same machinery.
• Note sharing or public links. Same reason.
• Refresh tokens. The default 30-day JWT is fine for a tutorial. Production apps would issue a short-lived access token plus a refresh flow.

How auth fits together

Three pieces work together so each user only sees their own notes:

Sign-in (who you are). Strapi's users-permissions plugin handles this for free. It ships a User content type, the register and login mutations, and a JWT issuer. Logging in returns { jwt, user }. Every later request sends Authorization: Bearer <jwt>, the plugin verifies the signature, and the user object is available as ctx.state.user for the rest of the request.

Authorization (what you can do). Sign-in tells the server who the caller is. Authorization decides what they can see or change. The rule across Part 4 is simple: only the owner. Reads filter to owner = ctx.state.user.id. Writes only succeed if the target note's owner.id matches the caller's id.

Ownership (the data that makes the check possible). We add owner as a manyToOne relation from Note to User. The server assigns it automatically when a note is created. From then on, the authorization check is just note.owner.id === user.id.

The diagram in the TL;DR shows the request path. The Document Service middleware does most of the work: it writes owner on create, and it adds an owner filter on read. The is-note-owner policy on the four GraphQL write mutations is the worked policy example for Part 4 and a second check before the resolver runs. After Step 6 the policy is technically redundant on the GraphQL side (the Document Service middleware already keeps a non-owner from loading the target note when Mutation.updateNote runs), but we keep it on purpose so you end up with a working policy file in your codebase to compare.

Step 1: Turn on sign-in and set the right permissions

users-permissions ships with every fresh Strapi project. Check that it is in package.json:

1"@strapi/plugin-users-permissions": "^5.x"

1.1 Check the JWT secret

The plugin signs JWTs with a value from the JWT_SECRET environment variable. Open .env and make sure it has a real value:

JWT_SECRET=<a-long-random-string>

If it is missing, generate one with openssl rand -hex 32 and paste it in. Anything 32 bytes or longer works. Keep it secret in production.

1.2 Lock down the Public role

After Part 2, anyone who can hit your server can read and write Notes and Tags. We need to close that. From now on, signed-out users can only register and log in. Everything else (listing notes, creating notes, updating notes) requires sign-in, and the rules in Steps 3 through 6 make sure each signed-in user only sees and edits their own data.

Open the admin UI at http://localhost:1337/admin → Settings → Users & Permissions Plugin → Roles → Public.

• Expand Note and uncheck everything. The Public role has no access to Notes.

• Expand Tag and leave find / findOne checked (tags are reference data, fine to expose). Uncheck create, update, delete.

• Expand Users-permissions → Auth and confirm register and callback are checked. They are usually enabled by default on the Public role; if either is unchecked, tick it now.

The other auth actions in that list (forgotPassword, resetPassword, emailConfirmation, sendEmailConfirmation, refresh, connect) are unused in this tutorial and can stay at whatever default Strapi shipped them with.

• Click Save.

Now go to Roles → Authenticated.

• Expand Note and check find, findOne, create, update. Leave delete unchecked.

• Expand Tag and check find and findOne. Anything else stays unchecked.

• Expand Users-permissions → User and confirm me is checked. It is enabled by default on the Authenticated role; the frontend will read it to display the signed-in user's name.

• Click Save.

1.3 Lock down the custom GraphQL mutations

In Part 2 we set auth: false on the three custom mutations (togglePin, archiveNote, duplicateNote) so they could be called without sign-in. That was fine while we had no users. Now we want them to require the same Authenticated role permissions you just set.

A bit of context on how the GraphQL plugin checks permissions:

• Shadow CRUD operations (Query.notes, Query.note, Mutation.createNote, etc.) come pre-wired to a permission name. The list query Query.notes checks for api::note.note.find; the single-item query Query.note checks for api::note.note.findOne; Mutation.createNote checks for api::note.note.create. Those names match the checkboxes in the admin UI's Roles screen, so ticking find, findOne, create, and update on the Authenticated role in 1.2 is enough to cover both list and detail reads plus the basic writes.
• Custom mutations (the three from Part 2) are not pre-wired to anything. The plugin has no way to guess togglePin is "really an update". Without a setting that tells it which permission to check, the only check left is "must be signed in", and any signed-in user passes.

If you only deleted the auth: false lines, anonymous calls would still be blocked, but any signed-in user could call those mutations. The role checkboxes would not gate them. That is inconsistent with the Shadow CRUD behavior we just relied on.

The fix is to tell each custom mutation which permission to check, using the auth.scope setting in resolversConfig. Pinning and archiving are kinds of update, so they reuse api::note.note.update. Duplicating creates a new row, so it reuses api::note.note.create.

Open src/extensions/graphql/mutations.ts, scroll to the resolversConfig block at the bottom, and swap the three auth: false entries:

1// before
2"Mutation.togglePin":     { auth: false },
3"Mutation.archiveNote":   { auth: false },
4"Mutation.duplicateNote": { auth: false },
5
6// after
7"Mutation.togglePin":     { auth: { scope: ["api::note.note.update"] } },
8"Mutation.archiveNote":   { auth: { scope: ["api::note.note.update"] } },
9"Mutation.duplicateNote": { auth: { scope: ["api::note.note.create"] } },

What this gives us:

• Anonymous requests are rejected before the resolver runs. No JWT, no entry.
• Signed-in users without the update permission are also rejected. If you ever add a read-only "Viewer" role, those users cannot call togglePin or archiveNote. The role checkbox decides.
• Signed-in users with update get through, then run into the is-note-owner policy from Step 4, which checks that the user actually owns the row they are touching.

Three checks before the resolver runs: signed in, has the right permission, owns the row. The first two come from auth.scope. The third is the policy in Step 4.

Same fix for the three custom queries. Part 2 also added searchNotes, noteStats, and notesByTag, and they have the same auth: false entries in src/extensions/graphql/queries.ts. Open that file, scroll to the resolversConfig block at the bottom, and swap:

1// before
2// "Query.searchArticles": { auth: false },
3"Query.searchNotes": { auth: false },
4"Query.noteStats":   { auth: false },
5"Query.notesByTag":  { auth: false },
6
7// after
8// "Query.searchArticles": { auth: false },
9"Query.searchNotes": { auth: { scope: ["api::note.note.find"] } },
10"Query.noteStats":   { auth: { scope: ["api::note.note.find"] } },
11"Query.notesByTag":  { auth: { scope: ["api::note.note.find"] } },

find is the right scope because all three are reads. Signed-out calls are now rejected; signed-in calls go through.

Leave Query.searchArticles alone — it is a Part 1 holdover for the Article content type and is not part of the Note ownership story.

What about scoping these to the current user? Partially. The Document Service middleware in Step 6 adds an owner filter to every findMany and findOne on api::note.note, so searchNotes and notesByTag would inherit the scope through filter merging. The genuine gaps are inside noteStats: it calls count() (the middleware skips it because the early return only matches findMany and findOne) and findMany on api::tag.tag (the middleware's uid check skips other content types entirely). On top of that, we want each owner check visible at the call site so a future reader of the resolver does not have to know the middleware is silently merging filters in. Step 6.3 adds an explicit owner filter to all three resolvers for those reasons.

Why reuse update instead of giving each mutation its own permission? You could add a custom controller method for each mutation so it shows up as a separate checkbox in the admin UI. That would let an "Editor" role toggle pin without being able to do a full update. We do not need that level of detail here. There is one Authenticated role, every signed-in user has the same permissions, so reusing update and create keeps the file count down. The Strapi GraphQL plugin docs cover the custom-permission variant if you need it.

1.4 Test in the Sandbox

Open the Apollo Sandbox at http://localhost:1337/graphql. Same Sandbox you used in Parts 1 and 2.

Check that the Public role cannot list notes. Paste this and run it (no headers):

1query {
2  notes {
3    documentId
4    title
5  }
6}

The response is Forbidden access. Good — signed-out reads are blocked before any middleware runs.

Register a user. We need an account to test with. Paste the mutation into the Operation editor first so the Sandbox knows the $input variable's type before you fill it in:

1mutation Register($input: UsersPermissionsRegisterInput!) {
2  register(input: $input) {
3    jwt
4    user {
5      username
6      email
7    }
8  }
9}

Then open the Variables panel at the bottom of the Sandbox and paste:

1{
2  "input": {
3    "username": "testuser",
4    "email": "testuser@example.com",
5    "password": "testuser"
6  }
7}

Click Register. The response has a JWT. Copy it — we will use it as a header in a moment.

Logging in. The JWT from register lasts 30 days, so you only register once per user. Every later session uses the login mutation. It takes identifier (username or email) and password and returns the same { jwt, user } shape:

1mutation Login($input: UsersPermissionsLoginInput!) {
2  login(input: $input) {
3    jwt
4    user {
5      username
6      email
7    }
8  }
9}

And the Variables panel with:

1{
2  "input": {
3    "identifier": "testuser",
4    "password": "testuser",
5    "provider": "local"
6  }
7}

provider: "local" is the built-in email-and-password provider. The input type requires it even though most apps never pass anything else. (OAuth flows would use "google", "github", etc.) Click Login and you get back a fresh JWT. The frontend in Step 8 calls this mutation when a returning user signs in.

List notes as testuser. Open the Headers panel and add a row. Header name is Authorization; value is the word Bearer, a space, and the JWT you copied earlier. No quotes, no backticks:

• Type the literal word Authorization into the Key field — eight letters, no surrounding characters. The backticks you see in inline-code formatting elsewhere in this post are markdown styling on a rendered page, not part of the value.
• In the Value field, type Bearer, then a single space, then paste the JWT.
• No quotes, no backticks, no leading or trailing whitespace.
• Make sure the checkbox on the left of the row is ticked. Sandbox keeps unticked rows around but does not send them.

Now re-run the notes query:

1query {
2  notes {
3    documentId
4    title
5  }
6}

This time the response is every note in the database.

That is the problem Part 4 fixes. testuser is signed in, but the API does not know about ownership yet, so they see everyone's notes. Steps 3 through 6 close that gap.

If you would rather use the terminal, the same three checks via curl:

# 1. Public role: Forbidden
curl -s -X POST http://localhost:1337/graphql \
  -H 'Content-Type: application/json' \
  -d '{"query":"{ notes { documentId title } }"}'
# -> {"errors":[{"message":"Forbidden access", ... }],"data":null}

# 2. Register testuser
curl -s -X POST http://localhost:1337/graphql \
  -H 'Content-Type: application/json' \
  -d '{"query":"mutation R($input: UsersPermissionsRegisterInput!) { register(input: $input) { jwt user { username email } } }","variables":{"input":{"username":"testuser","email":"testuser@example.com","password":"testuser"}}}'
# -> {"data":{"register":{"jwt":"<JWT>", ... }}}

# 3. Signed-in read: every note in the database
JWT="<paste-the-token>"
curl -s -X POST http://localhost:1337/graphql \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $JWT" \
  -d '{"query":"{ notes { documentId title } }"}'
# -> {"data":{"notes":[ ... every note in the database ... ]}}

Step 2: Add the owner relation to Note

Open the admin UI → Content-Type Builder → Note → Add another field:

1. Pick Relation.
2. Leave Note on the left.
3. On the right, pick User (from: users-permissions).
4. Choose manyToOne: one user has many notes, each note has one owner.
5. Name the field owner on the Note side and notes on the User side.
6. Click Finish, then Save. Strapi restarts.

After restart, the Note schema has:

1{
2  "owner": {
3    "type": "relation",
4    "relation": "manyToOne",
5    "target": "plugin::users-permissions.user",
6    "inversedBy": "notes"
7  }
8}

The relation is nullable by default. Step 3 fixes that by stamping an owner automatically on every new note.

2.1 Backfill existing notes

The notes you seeded in Part 3 do not have an owner yet. The cleanest fix is a one-shot script that bootstraps Strapi, assigns every existing note to testuser, and exits. From server/, create scripts/backfill-owner.js:

1// server/scripts/backfill-owner.js
2"use strict";
3
4async function main() {
5  const { createStrapi, compileStrapi } = require("@strapi/strapi");
6  const app = await createStrapi(await compileStrapi()).load();
7  app.log.level = "error";
8
9  const testuser = await app
10    .documents("plugin::users-permissions.user")
11    .findFirst({ filters: { username: "testuser" } });
12  if (!testuser) {
13    throw new Error('No user with username "testuser". Register one first.');
14  }
15
16  const notes = await app
17    .documents("api::note.note")
18    .findMany({ pagination: { pageSize: 100 } });
19
20  for (const note of notes) {
21    await app.documents("api::note.note").update({
22      documentId: note.documentId,
23      data: { owner: testuser.id },
24    });
25  }
26
27  console.log(
28    `Backfilled ${notes.length} notes to owner ${testuser.username}.`,
29  );
30
31  await app.destroy();
32  process.exit(0);
33}
34
35main().catch((err) => {
36  console.error(err);
37  process.exit(1);
38});

Run it with node scripts/backfill-owner.js. The script boots a Strapi instance, runs the backfill against the Document Service, and exits. Or, if you prefer a GUI-based approach, open the admin UI's Content Manager and assign the owner relation on each note by hand.

Every existing note now has testuser as its owner. New notes get one automatically once Step 3 is in place.

Step 3: Set the owner automatically on every new note

Clients should not be able to pick an owner. Every signed-in request has the user available as ctx.state.user; we want to use that to set owner on a new note no matter how the request comes in.

Strapi gives you four reasonable places to do this. Pick once and the same answer works for any "set this field on create" rule you hit later (createdBy, tenantId, defaults).

Option A: Content-type lifecycle hook

A lifecycles.ts file in src/api/note/content-types/note/ exporting a beforeCreate function that modifies event.params.data before the row is written. This was the standard answer in Strapi v4.

Strapi v5 still supports it, but Content lifecycle management and When to use lifecycle hooks in Strapi explain why it is no longer the recommended option:

• It is buried inside a content-type folder, so a new developer has to know to look for it.
• It runs inside the database transaction, which makes error handling harder.
• There is no way to opt a single call out of the rule.

Skipping.

Option B: GraphQL middleware on Mutation.createNote

A resolversConfig middleware that reads context.state.user and writes args.data.owner before the resolver runs. Same setup as the cap-page-size policy from Part 2:

1// hypothetical entry in src/extensions/graphql/middlewares-and-policies.ts
2"Mutation.createNote": {
3  middlewares: [
4    async (next, parent, args, context, info) => {
5      const user = context?.state?.user;
6      if (user?.id) {
7        args.data = { ...(args.data ?? {}), owner: user.id };
8      }
9      return next(parent, args, context, info);
10    },
11  ],
12},

Clean for GraphQL. But REST POST /api/notes does not run through this middleware. A client that sends {"data": {"owner": 99, "title": "..."}} to the REST endpoint writes owner: 99 straight to the database. Same half-coverage problem we hit in Part 2.

Option C: REST route middleware on the Note router

The mirror of Option B. A Koa middleware on src/api/note/routes/note.ts that writes ctx.request.body.data.owner before the controller runs. Covers POST /api/notes, does nothing for GraphQL. The gap is now on the other side.

You can do Options B and C together and cover both APIs, but then the rule lives in two files with two different argument names (args.data.owner vs ctx.request.body.data.owner). One side gets updated, the other does not, and the two slowly drift apart.

Option D: Document Service middleware (what we use)

A Document Service middleware registered in register() runs whenever any code calls strapi.documents("api::note.note").create(...). The Document Service sits below both REST and GraphQL: every controller and every GraphQL resolver eventually calls into it. So a rule here covers REST, GraphQL, custom resolvers (including Mutation.duplicateNote, which calls strapi.documents("api::note.note").create(...) directly), the seed script, and the admin UI, all from one file.

Coverage

Why Option D

Two reasons.

REST is always on in Strapi. Even if your code only calls GraphQL, the REST routes still exist and accept requests. From the outside, every "GraphQL-only" Strapi app is also a REST API. A rule about how data gets written has to apply to both. Picking Option B and saying "we only use GraphQL" creates a hole that opens the first time a developer, a script, or an attacker sends a POST /api/notes request.

Only Option D sits below both APIs. Options B and C run inside one API's request handling: B inside the GraphQL resolver wrapper, C inside the Koa router. Option D runs inside the Document Service, which both APIs call into for every read and every write. A rule placed there cannot be bypassed by a new code path, because anything that touches Note data has to go through strapi.documents("api::note.note").<action>(...).

So data rules (like "set this field automatically on create") go in Document Service middlewares. Step 3 puts owner-stamping there. Step 6 will add a second clause to the same middleware for read-scoping. One file holds both.

Open server/src/index.ts and register the middleware in register(). The Document Service middlewares docs say middlewares should be registered in the register() lifecycle, not bootstrap(), so the rule is in place before any plugin's bootstrap runs.

1// server/src/index.ts
2import type { Core } from "@strapi/strapi";
3import registerGraphQLExtensions from "./extensions/graphql";
4
5export default {
6  register({ strapi }: { strapi: Core.Strapi }) {
7    registerGraphQLExtensions(strapi);
8
9    // Stamp the authenticated user as `owner` on every new Note. Runs at
10    // the Document Service layer, so it applies to REST, GraphQL, custom
11    // resolvers, and seed scripts uniformly.
12    strapi.documents.use(async (context, next) => {
13      if (context.uid !== "api::note.note") return next();
14      if (context.action !== "create") return next();
15
16      const requestCtx = strapi.requestContext.get();
17      const user = requestCtx?.state?.user;
18      if (user?.id) {
19        context.params.data = {
20          ...(context.params.data ?? {}),
21          owner: user.id,
22        } as typeof context.params.data;
23      }
24
25      return next();
26    });
27  },
28
29  bootstrap() {
30    // Bootstrap intentionally empty for now. Step 6 leaves it that way too.
31  },
32};

Three things to notice:

• The middleware overwrites whatever the client sent. Even if a client sends data: { owner: 99 } to createNote or POST /api/notes, the middleware runs at the data layer and writes owner: <signed-in user's id> instead. The client's 99 is dropped. The "I'll claim to be user 99" hole closes here, below the place either API hands the data off.
• A signed-out create produces a note with no owner. That is the right outcome, and Step 1 already blocks signed-out create through the role permissions before we ever reach the middleware. Two checks, one after the other.
• Reads are unaffected. The early return next() when context.action !== "create" makes the middleware skip on every other action (findMany, findOne, update, delete). Step 6 adds a second clause to the same middleware for read-scoping.

Restart Strapi and test in the Sandbox.

In the Sandbox at http://localhost:1337/graphql, make sure the Authorization header from Step 1.4 is still ticked. The two operations below show the middleware did its job without us having to read the owner field back through GraphQL.

Paste both into the Operation editor at once. The Sandbox supports multiple named operations in one document and shows a dropdown next to the run button:

1query Me {
2  me {
3    id
4    username
5  }
6}
7
8mutation CreateNote {
9  createNote(
10    data: { title: "testuser owns me", content: "yo", pinned: false }
11  ) {
12    documentId
13    title
14  }
15}

Run Me first. The response shows username: "testuser". That is who the next mutation will be authenticated as, and the value the upcoming MyNotes query will filter by.

Then run CreateNote. The response contains a documentId and the title you sent. The data input did not include an owner field anywhere; the Document Service middleware stamped it server-side before the resolver wrote the row.

List the current user's notes

We need one more verification: read back the notes belonging to the signed-in user and confirm the new note shows up.

Try this query first. Add this to the Sandbox document and run it (Authorization header still set):

1query BrokenMyNotes($username: String!) {
2  notes(filters: { owner: { username: { eq: $username } } }) {
3    documentId
4    title
5  }
6}

Variables:

1{ "username": "testuser" }

The response will be an error:

1{
2  "errors": [
3    {
4      "message": "Invalid key owner",
5      "path": ["notes"],
6      "extensions": {
7        "code": "BAD_USER_INPUT",
8        "error": {
9          "name": "ValidationError",
10          "details": {
11            "key": "owner",
12            "path": "owner",
13            "source": "query",
14            "param": "filters"
15          }
16        }
17      }
18    }
19  ],
20  "data": null
21}

Reading the stack trace from top to bottom tells you exactly what happened:

• ValidationError: Invalid key owner: Strapi rejects owner as a key the caller is not allowed to use in this position (the filters argument).
• handleRegularRelation in @strapi/utils/.../throw-restricted-relations.js:87: this function walks every relation named in filters and checks whether the caller has read permission on the related content type. owner points at User. The Authenticated role has no permissions on User. The check fails.
• validateFilters → validateQuery → findMany in the GraphQL plugin: the rejection happens inside the resolver. The Part 2 soft-delete middleware has already run by this point, and the database query has not yet been built.
• resolversConfig.Query.notes.middlewares at the bottom: our Part 2 soft-delete middleware. It is innocent here; the error fires after it.

The takeaway: filtering on a relation requires read permission on the related content type, even if the response never selects any field from that relation. Strapi enforces this during query validation, before the resolver runs the database query.

The first instinct: grant the permission and move on. It is right there in the admin UI. Open http://localhost:1337/admin → Settings → Users & Permissions Plugin → Roles → Authenticated → expand Users-permissions → User → tick find. Save.

Re-run BrokenMyNotes in the Sandbox. The error is gone; the response is the array of notes whose owner.username equals testuser. Filter validated, query succeeded.

Why this is the wrong fix. That same find permission also unblocks the standard list endpoint for the User content type. With testuser still authenticated, run this in the Sandbox:

Heads-up before running this. If testuser is the only account you have registered so far, the response below will only contain one row and the privacy leak will not be obvious. Register a second user first (any username and email; reuse the Register mutation block from Step 1.4 with new variables, or sign in to the Sandbox with testuser2 / testuser2 / testuser2@example.com) so the query has more than one row to leak.

1query AllUsers {
2  usersPermissionsUsers {
3    documentId
4    username
5    email
6  }
7}

The response is every user in the database: testuser, testuser2, anyone else who registered, all their usernames and emails handed to any signed-in caller. The admin UI does not have a "let me filter by user, but don't let me list users" toggle. The find permission covers both. That is a real privacy leak.

Go back to the admin UI and uncheck find on User for the Authenticated role. Save. We need an approach that does not require the permission at all.

The right approach: a custom Query.myNotes resolver. It reads the user id from the JWT server-side, then filters using the Document Service. The Document Service builds the query directly against the database, so it does not run through the GraphQL validation that rejected our owner filter above. The client never names owner (the query takes no arguments), so the validation has nothing to reject. The User content type stays fully closed; signed-in callers cannot list users.

Step 1: define a payload type and add the field. Open server/src/extensions/graphql/queries.ts. We want myNotes to return both the signed-in user (so the frontend can render "logged in as ") and the list of notes, in one round trip.

Add two new object types as siblings to the existing TagCount and NoteStats declarations at the top of the types array. The first, MyNotesUser, is a deliberately small view of the user that only exposes id and username (the only two fields a list page actually needs). The second, MyNotesPayload, wraps that user with the notes array:

1nexus.objectType({
2  name: "MyNotesUser",
3  definition(t) {
4    t.nonNull.id("id");
5    t.nonNull.string("username");
6  },
7}),
8nexus.objectType({
9  name: "MyNotesPayload",
10  definition(t) {
11    t.nonNull.field("user", { type: "MyNotesUser" });
12    t.nonNull.list.nonNull.field("notes", { type: "Note" });
13  },
14}),

Why a custom MyNotesUser instead of reusing the built-in UsersPermissionsMe type? UsersPermissionsMe exposes id, documentId, username, email, confirmed, blocked, and role. If we returned that type from myNotes, a client could ask for any of those fields by writing them in their query (for example myNotes { user { email role { name } } }), and the server would return them. Since this query is only meant to power a list page that needs username and nothing else, we say so in the schema itself. MyNotesUser has two fields, and those are the only two fields a client of myNotes can ever ask for. The schema is the contract; relying on the client to "select less" is not.

If a page genuinely needs email, role, or other user fields (think a profile or settings page), it calls Query.me directly on that page. Two queries, two purposes, each one returning only the fields its page actually renders.

Then inside the existing nexus.extendType({ type: "Query", definition(t) { ... } }) block, after the notesByTag field, add the resolver:

1t.field("myNotes", {
2  type: nexus.nonNull("MyNotesPayload"),
3  async resolve(_parent: unknown, _args: unknown, ctx: any) {
4    const user = ctx?.state?.user;
5    if (!user?.id) {
6      // Should never reach here because of the `auth.scope` config below,
7      // but a tutorial-grade fallback. Return an empty payload shape that
8      // the frontend can render without special cases.
9      return { user: { id: 0, username: "" }, notes: [] };
10    }
11    const notes = await strapi.documents("api::note.note").findMany({
12      filters: { owner: { id: user.id } },
13      sort: ["pinned:desc", "updatedAt:desc"],
14      populate: ["tags"],
15    });
16    return {
17      user: { id: user.id, username: user.username },
18      notes,
19    };
20  },
21});

The resolver explicitly picks only id and username off ctx.state.user and assembles the slim payload. Even if a future change adds new fields to MyNotesUser, the resolver only sends what it explicitly chooses to.

Step 2: tie it to the role permission. In the same file's resolversConfig block at the bottom, add:

1"Query.myNotes": { auth: { scope: ["api::note.note.find"] } },

This requires the caller to have find on Note (which the Authenticated role already has from Step 1.2). Anonymous calls are rejected before the resolver runs.

Step 3: save and let Strapi restart. The dev server picks up the change automatically.

Step 4: run the working query. Replace the broken BrokenMyNotes operation in the Sandbox with:

1query MyNotes {
2  myNotes {
3    user {
4      id
5      username
6    }
7    notes {
8      documentId
9      title
10      pinned
11    }
12  }
13}

The response contains the current user's id and username plus the array of notes owned by them, including the one CreateNote just created. Run it as testuser2 (different JWT) and the user block changes to testuser2 and the notes array contains only their notes.

What this resolver gets right:

• The user id stays on the server. The client never sees, stores, or passes it. The JWT is the source of truth, and the server is the only thing that reads it.
• No new role permissions needed. The User content type stays closed, and no usernames or emails leak to other signed-in users.
• The schema only exposes two fields on the user. A client of myNotes can only ever ask for id and username. There is no way to write a query that pulls back email or role from this endpoint, even if a future change to the resolver accidentally returns more.
• One endpoint, zero client logic. A page that wants "the signed-in user's notes" calls myNotes and is done. There is no path where the client could pass a wrong id, because the client does not pass an id at all.

Looking ahead: Document Service middleware vs custom resolver. A custom resolver like myNotes is the right tool when you want a named endpoint that the schema makes obvious (a query called myNotes is a clear contract: "the current user's notes"). A Document Service middleware is the right tool when you want the existing endpoint to behave correctly without any per-call work from the caller (the bare notes query should always be scoped, no matter who calls it from where). Step 6 adds a Document Service middleware so the bare notes query also returns only the caller's notes, on both REST and GraphQL. Neither approach replaces the other; we use both in the final design.

Step 4: A first cut at the is-note-owner policy

In Part 2 we said the real authorization example would land in Part 4. Here it is.

Only the note's owner should be able to update it. That means gating these four mutations, all of which take a note's documentId as input:

• Mutation.updateNote
• Mutation.togglePin
• Mutation.archiveNote
• Mutation.duplicateNote

Mutation.createNote is intentionally excluded; ownership is set there by the Document Service middleware from Step 3, not checked.

Create server/src/policies/is-note-owner.ts:

1// server/src/policies/is-note-owner.ts
2import type { Core } from "@strapi/strapi";
3
4type PolicyContext = {
5  args?: { documentId?: string };
6  state?: { user?: { id?: number | string } };
7  context?: { state?: { user?: { id?: number | string } } };
8};
9
10const isNoteOwner = async (
11  policyContext: PolicyContext,
12  _config: unknown,
13  { strapi }: { strapi: Core.Strapi },
14): Promise<boolean> => {
15  const user =
16    policyContext?.state?.user ?? policyContext?.context?.state?.user;
17  if (!user?.id) {
18    strapi.log.warn("is-note-owner: rejected, no authenticated user.");
19    return false;
20  }
21
22  const documentId = policyContext?.args?.documentId;
23  if (!documentId) {
24    strapi.log.warn("is-note-owner: rejected, no documentId in args.");
25    return false;
26  }
27
28  const note = await strapi
29    .documents("api::note.note")
30    .findOne({ documentId, populate: ["owner"] });
31
32  if (!note) return false;
33  if (note.owner?.id === user.id) return true;
34
35  strapi.log.warn(
36    `is-note-owner: rejected, user ${user.id} is not the owner of note ${documentId}.`,
37  );
38  return false;
39};
40
41export default isNoteOwner;

Now wire the policy onto the four write mutations.

Open server/src/extensions/graphql/middlewares-and-policies.ts and find the resolversConfig block. It already has entries for Query.notes and Query.note from Part 2. Add four new sibling keys inside the same object, right after the Query.note block:

1return {
2  resolversConfig: {
3    "Query.notes": {
4      // ... unchanged from Part 2 ...
5    },
6    "Query.note": {
7      // ... unchanged from Part 2 ...
8    },
9
10    // NEW: per-row ownership check on every Note write
11    "Mutation.updateNote": { policies: ["global::is-note-owner"] },
12    "Mutation.togglePin": { policies: ["global::is-note-owner"] },
13    "Mutation.archiveNote": { policies: ["global::is-note-owner"] },
14    "Mutation.duplicateNote": { policies: ["global::is-note-owner"] },
15  },
16};

A few things to know:

• The four new keys go inside the same resolversConfig object as Query.notes and Query.note. Same indentation, separated by commas.
• Mutation.createNote is not in the list. Ownership is set on create by the Document Service middleware from Step 3, not checked.
• All four mutations take documentId as an argument. That is what the policy reads to load the target row.
• The string "global::is-note-owner" matches the filename src/policies/is-note-owner.ts you wrote above. Strapi auto-registers files in src/policies/ under that prefix; if the file is missing or misnamed, Strapi will fail to start with Cannot find policy is-note-owner.

Test the policy in the Sandbox

We need a second user to play the part of "not the owner." If you already registered one earlier (for the privacy-leak demonstration in Step 3), reuse the Login mutation block from Step 1.4 to grab a fresh JWT for that account, then jump down to the TogglePin test. Otherwise, register testuser2 now in the Sandbox:

1mutation Register($input: UsersPermissionsRegisterInput!) {
2  register(input: $input) {
3    jwt
4    user {
5      username
6    }
7  }
8}

Variables:

1{
2  "input": {
3    "username": "testuser2",
4    "email": "testuser2@example.com",
5    "password": "testuser2"
6  }
7}

Copy testuser2's JWT.

Now grab any existing note's documentId (they all belong to testuser after the backfill in Step 2.1). Swap the Authorization header in the Sandbox to Bearer <testuser2-jwt> and try to pin testuser's note:

1mutation TogglePin($id: ID!) {
2  togglePin(documentId: $id) {
3    documentId
4    pinned
5  }
6}

Variables:

1{ "id": "<paste-note-documentId>" }

The response is Policy Failed.

Swap the header back to testuser's JWT and run the same mutation. Now it succeeds and the note flips between pinned and unpinned.

Step 5: The REST leak

Anything in resolversConfig only runs for GraphQL. The same gap we hit in Part 2 with the soft-delete and page-cap rules. The is-note-owner policy does nothing for REST, and testuser2 can prove it by editing testuser's note over REST:

TESTUSER2_JWT="<testuser2-jwt>"
NOTE_ID="<a-note-documentId>"
curl -s -X PUT "http://localhost:1337/api/notes/$NOTE_ID" \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $TESTUSER2_JWT" \
  -d '{"data":{"title":"testuser2 was here"}}'
# -> 200 OK, title updated.

That is a real bug. The policy gates GraphQL but not REST. One rule, half-covered.

Restore the title before moving on:

TESTUSER_JWT="<testuser-jwt>"
curl -s -X PUT "http://localhost:1337/api/notes/$NOTE_ID" \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $TESTUSER_JWT" \
  -d '{"data":{"title":"<original title>"}}'

Two reasonable ways to close the gap:

1. Mirror the rule on REST. Add a route-level middleware on src/api/note/routes/note.ts (same setup Part 2 used for soft-delete) plus a small policy on the route's find, findOne, update, and delete actions that loads the target note and rejects when the owner does not match. The rule now lives in two files: a resolversConfig policy for GraphQL and a route middleware/policy for REST. Both have to be kept in sync.
2. Move the rule below both APIs into a Document Service middleware. Step 3 already uses one to write owner on create. We extend that same middleware with a second clause: it watches every findMany and findOne on api::note.note and adds owner = ctx.state.user.id to params.filters. REST controllers and GraphQL Shadow CRUD resolvers both eventually call strapi.documents("api::note.note").<action>(...), so a rule placed there fires for both. One file, both APIs.

The next step builds option 2.

Why not option 1? It works, but it puts the same rule in two files. A future change updates the GraphQL side and forgets the REST side, and the same kind of leak we just demonstrated comes back. Option 2 puts the rule one layer deeper, where REST and GraphQL share the same code path: any caller that ends up at strapi.documents("api::note.note").findOne(...) goes through the rule, with no extra wiring. We took a similar position in Part 2's "GraphQL-only vs. both APIs" section. Part 4's ownership rule is the case where that one-rule-one-file approach earns its keep, because the rule has to read ctx.state.user and behave the same way for REST and GraphQL.

Step 6: A Document Service middleware for both APIs

The owner-stamping middleware from Step 3 already lives in src/index.ts register(). We add two more strapi.documents.use(...) clauses in the same file, sitting right after it.

The three clauses together do three things:

1. Stamp owner on every new note (the existing Step 3 clause).
2. Filter findMany and findOne to the signed-in user. Both REST GET /api/notes / /:id and GraphQL Query.notes / Query.note go through the Document Service, so one rule covers both.
3. Block update and delete for non-owners. A pre-flight findOne (which inherits the read filter from clause 2) returns null for non-owners; we throw NotFoundError. REST PUT / DELETE /api/notes/:id is covered the same way GraphQL updateNote / deleteNote is.

The rules live at the Document Service layer, so REST and GraphQL share them. There is no per-resolver attachment on the GraphQL side and no route-level middleware on the REST side. One file, both APIs, both verbs.

6.1 Extend the Document Service middleware

Open server/src/index.ts and replace its contents:

1// server/src/index.ts
2import type { Core } from "@strapi/strapi";
3import { errors } from "@strapi/utils";
4import registerGraphQLExtensions from "./extensions/graphql";
5
6export default {
7  register({ strapi }: { strapi: Core.Strapi }) {
8    registerGraphQLExtensions(strapi);
9
10    // From Step 3: stamp `owner` on every new note.
11    strapi.documents.use(async (context, next) => {
12      if (context.uid !== "api::note.note") return next();
13      if (context.action !== "create") return next();
14
15      const requestCtx = strapi.requestContext.get();
16      const user = requestCtx?.state?.user;
17      if (user?.id) {
18        context.params.data = {
19          ...(context.params.data ?? {}),
20          owner: user.id,
21        } as typeof context.params.data;
22      }
23
24      return next();
25    });
26
27    // Scope every Note read to the signed-in user.
28    strapi.documents.use(async (context, next) => {
29      if (context.uid !== "api::note.note") return next();
30      if (context.action !== "findMany" && context.action !== "findOne") {
31        return next();
32      }
33
34      const requestCtx = strapi.requestContext.get();
35      const user = requestCtx?.state?.user;
36      const existingFilters = (context.params.filters ?? {}) as Record<
37        string,
38        unknown
39      >;
40
41      if (!user?.id) {
42        // No signed-in user: force the filter to match nothing.
43        // -1 is never a valid User id, so the query returns zero rows
44        // instead of falling through to "return everything."
45        context.params.filters = {
46          ...existingFilters,
47          owner: { id: { $eq: -1 } },
48        } as typeof context.params.filters;
49        return next();
50      }
51
52      context.params.filters = {
53        ...existingFilters,
54        owner: { id: { $eq: user.id } },
55      } as typeof context.params.filters;
56      return next();
57    });
58
59    // Gate every Note write (update, delete) for non-owners.
60    // The pre-flight findOne goes back through the Document Service
61    // and inherits the read filter above, so a non-owner sees null
62    // for the lookup. We throw NotFoundError to match the read
63    // posture (no existence leak via 403).
64    strapi.documents.use(async (context, next) => {
65      if (context.uid !== "api::note.note") return next();
66      if (context.action !== "update" && context.action !== "delete") {
67        return next();
68      }
69
70      const documentId = (context.params as { documentId?: string })
71        .documentId;
72      if (!documentId) return next();
73
74      const existing = await strapi
75        .documents("api::note.note")
76        .findOne({ documentId });
77      if (!existing) {
78        throw new errors.NotFoundError("Note not found.");
79      }
80
81      return next();
82    });
83  },
84
85  bootstrap() {},
86};

What this code does.

On reads (findMany and findOne):

• It looks up who is signed in. Inside this middleware we do not get the request directly, so we ask Strapi for it with strapi.requestContext.get(). If testuser is signed in, user.id is testuser's id. If nobody is signed in, user is undefined.
• If nobody is signed in, the query returns zero notes. We add the filter owner.id = -1 to the query. No user has id -1, so the database finds nothing. We do not throw an error here because the route in Step 1 already rejects requests without a JWT. This is just a backstop in case anything slips past it.
• The caller cannot ask for someone else's notes. If a caller sends notes(filters: { owner: { id: { eq: 99 } } }) hoping to read user 99's notes, we copy their filter into ours, then write our own owner value on top. Our value wins, so the database only returns notes owned by the signed-in user.

On writes (update and delete):

• It checks ownership by looking the note up first. Before the update or delete runs, we call strapi.documents("api::note.note").findOne({ documentId }). That lookup goes through the same middleware, so the read code above runs on it and adds owner = signed-in user to the query. If testuser2 tries to update one of testuser's notes, the lookup returns null, and we throw. We never write the ownership check twice; the write code borrows it from the read code.
• It throws NotFoundError, not ForbiddenError. When testuser2 reads testuser's note, the read code above already returns nothing, as if the note does not exist. We want the write to look the same. ForbiddenError would tell testuser2 "this note exists, you just cannot touch it," which tells them something they should not know.
• It only runs on update and delete. create is already handled by the owner-stamping code at the top of the file. Plain reads are handled by the read code above. count is handled inside noteStats in Step 6.3.

Heads-up: this fires on the admin UI too. The Document Service sits below every API surface, including the Content Manager in http://localhost:1337/admin. When you open Note in the admin UI, the request runs as your admin user (a row in admin::user), and strapi.requestContext.get()?.state?.user returns that admin user. The owner filter then becomes owner.id = <admin-user-id>, but owner points at plugin::users-permissions.user, which is a different content type. The two id spaces happen to overlap by coincidence, so the Content Manager's Note list will look mostly empty (or show the wrong rows) until you log in as the end-user whose id matches an admin id. If you need the Content Manager to keep working as an admin, branch on requestCtx?.state?.auth?.strategy and return next() early when the strategy is the admin one ("admin") so admins keep their full view; the rule we wrote above only applies to the public REST and GraphQL surfaces. The tutorial leaves the simpler version in place because the demo never opens the Note Content Manager again, and an admin caveat hides the rule we just spent the section building.

6.2 Test it

Restart Strapi, then exercise reads and writes on both APIs.

GraphQL reads. With testuser's JWT in the Authorization header, run:

1query {
2  notes {
3    documentId
4  }
5}

You see the notes testuser owns. Swap the Authorization header to testuser2's JWT and re-run. testuser2 sees an empty list (or only the notes they created themselves).

GraphQL writes. Try to update one of testuser's notes as testuser2:

1mutation UpdateNote($id: ID!, $data: NoteInput!) {
2  updateNote(documentId: $id, data: $data) {
3    documentId
4    title
5  }
6}

Variables:

1{
2  "id": "<testuser-note-documentId>",
3  "data": { "title": "testuser2 was here" }
4}

The response is Policy Failed. The is-note-owner policy from Step 4 rejects the call before the resolver runs. Even if the policy were absent, the write-gate clause would now throw NotFoundError from the same lookup.

REST reads. Same shape, different surface:

TESTUSER_JWT="<testuser-jwt>"
TESTUSER2_JWT="<testuser2-jwt>"

curl -s -H "Authorization: Bearer $TESTUSER_JWT" "http://localhost:1337/api/notes" | jq '.data | length'
# -> 9

curl -s -H "Authorization: Bearer $TESTUSER2_JWT" "http://localhost:1337/api/notes" | jq '.data | length'
# -> 0

REST writes. This is what the write-gate clause is for. Pick any of testuser's note documentIds and try to PUT to it as testuser2:

NOTE_ID="<testuser-note-documentId>"
curl -s -X PUT "http://localhost:1337/api/notes/$NOTE_ID" \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $TESTUSER2_JWT" \
  -d '{"data":{"title":"testuser2 was here"}}'
# -> {"data":null,"error":{"status":404,"name":"NotFoundError", ... }}

testuser2's PUT returns 404. testuser's own PUT on the same note returns 200 and updates the title.

curl -s -X PUT "http://localhost:1337/api/notes/$NOTE_ID" \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $TESTUSER_JWT" \
  -d '{"data":{"title":"testuser updated this"}}'
# -> 200, title now "testuser updated this"

One file (src/index.ts), three clauses, both APIs, both verbs. Reads scoped, writes gated, ownership stamped on create. The intro promise — a Document Service middleware that closes both APIs from one file — holds.

6.3 Scope the three custom queries to the owner

The Document Service middleware covers Query.notes, Query.note, and the REST find/findOne, because all four end up calling strapi.documents("api::note.note").findMany(...) or findOne(...). The three custom queries from Part 2 (searchNotes, noteStats, notesByTag) are different:

• searchNotes and notesByTag call strapi.documents("api::note.note").findMany(...) themselves, with their own filter object. The middleware's owner clause does merge in, but it is harder to read what the final filter is from inside the resolver, and any future change to the resolver could clobber it. Adding the owner clause inline makes the rule visible at the call site.
• noteStats calls count() three times (which the middleware does not match because the early return only allows findMany and findOne) and then findMany on api::tag.tag (which is a different content type, so the middleware's uid check skips it entirely).

Step 1.3 already made all three queries require sign-in. We now add an explicit owner filter inside each resolver so each signed-in user only sees their own data.

The same pattern in all three: read ctx.state.user.id and add owner: { id: user.id } to the filter. Each resolver now takes ctx as its third argument.

Open server/src/extensions/graphql/queries.ts.

searchNotes. Replace the resolver with:

1t.list.field("searchNotes", {
2  type: nexus.nonNull("Note"),
3  args: {
4    query: nexus.nonNull(nexus.stringArg()),
5    includeArchived: nexus.booleanArg({ default: false }),
6  },
7  async resolve(
8    _parent: unknown,
9    { query, includeArchived }: { query: string; includeArchived: boolean },
10    ctx: any,
11  ) {
12    const user = ctx?.state?.user;
13    if (!user?.id) return [];
14    const where: any = {
15      title: { $containsi: query },
16      owner: { id: user.id },
17    };
18    if (!includeArchived) where.archived = false;
19    return strapi.documents("api::note.note").findMany({
20      filters: where,
21      populate: ["tags"],
22      sort: ["pinned:desc", "updatedAt:desc"],
23    });
24  },
25});

noteStats. Each count and findMany call gets the owner filter:

1t.nonNull.field("noteStats", {
2  type: "NoteStats",
3  async resolve(_parent: unknown, _args: unknown, ctx: any) {
4    const user = ctx?.state?.user;
5    if (!user?.id) return { total: 0, pinned: 0, archived: 0, byTag: [] };
6    const owner = { id: user.id };
7    const [total, pinned, archived, tags] = await Promise.all([
8      strapi.documents("api::note.note").count({ filters: { owner } }),
9      strapi.documents("api::note.note").count({
10        filters: { pinned: true, owner },
11      }),
12      strapi.documents("api::note.note").count({
13        filters: { archived: true, owner },
14      }),
15      strapi.documents("api::tag.tag").findMany({
16        populate: { notes: { filters: { owner } } },
17        sort: ["name:asc"],
18      }),
19    ]);
20
21    const byTag = tags
22      .map((tag: any) => ({
23        slug: tag.slug,
24        name: tag.name,
25        count: Array.isArray(tag.notes) ? tag.notes.length : 0,
26      }))
27      .sort((a, b) => b.count - a.count || a.name.localeCompare(b.name));
28
29    return { total, pinned, archived, byTag };
30  },
31});

The line worth pointing at is populate: { notes: { filters: { owner } } }. Every user can see every tag (tags are shared), but the number of notes on each tag has to be scoped to the current user. The Document Service lets us filter populated relations inline, so we add { owner } to the populate. Each user sees their own per-tag counts, never anyone else's.

notesByTag. Add owner to the existing filter object:

1t.list.field("notesByTag", {
2  type: nexus.nonNull("Note"),
3  args: { slug: nexus.nonNull(nexus.stringArg()) },
4  async resolve(_parent: unknown, { slug }: { slug: string }, ctx: any) {
5    const user = ctx?.state?.user;
6    if (!user?.id) return [];
7    return strapi.documents("api::note.note").findMany({
8      filters: {
9        archived: false,
10        tags: { slug: { $eq: slug } },
11        owner: { id: user.id },
12      },
13      populate: ["tags"],
14      sort: ["pinned:desc", "updatedAt:desc"],
15    });
16  },
17});

Save. Strapi auto-restarts.

Test in the Sandbox. With testuser's JWT in the Authorization header, run each one:

1query {
2  searchNotes(query: "GraphQL") {
3    documentId
4    title
5  }
6}

1query {
2  noteStats {
3    total
4    pinned
5    archived
6    byTag {
7      slug
8      name
9      count
10    }
11  }
12}

1query {
2  notesByTag(slug: "ideas") {
3    documentId
4    title
5  }
6}

You only see testuser's notes. Swap the Authorization header to testuser2's JWT and re-run the same queries. testuser2 only sees their own notes.

Why repeat the owner filter inline in each resolver instead of pulling it into a helper? A withOwner(filters, ctx) helper looks like the obvious refactor, but each resolver here is short, and the helper would barely save any lines. Writing the owner filter inline also keeps the rule visible at every read site, so a future reader of the resolver does not have to know that a helper exists. With twenty custom queries the helper would pay off; with three, inline is plainer.

6.4 Should we keep the GraphQL policy?

After Step 6, the is-note-owner policy from Step 4 is technically redundant. The Document Service middleware adds an owner filter to every findOne, so when Mutation.updateNote (or togglePin, archiveNote, duplicateNote) calls strapi.documents("api::note.note").findOne(...) to load the target note, a non-owner gets back nothing. The mutation fails before the policy would have had a chance to reject it.

The policy is still useful as a second check and as a worked example of how policies are written. Two reasonable choices:

• Keep it. A second check, run before the resolver. If someone disables or breaks the Document Service middleware by accident, GraphQL writes are still gated. The cost is that the same rule lives in two places.
• Delete it. One source of truth. Trust the Document Service middleware to do the work.

We keep the policy in this tutorial. The file is small, the overlap is on purpose, and the codebase ends up with two patterns side by side: a Document Service middleware doing the main work, plus a policyContext-based policy as a worked example.

Step 7: Test the backend contract

The starter's test script already covers Part 2. Add a section at the bottom that covers ownership. Open starter-template/scripts/test-graphql.mjs (and the canonical copy in server/scripts/test-graphql.mjs) and append:

1// 9. Ownership: two-user isolation
2section("Ownership: two-user isolation");
3
4const testuserJwt = (
5  await gql(
6    `mutation { register(input: { username: "testuser-${Date.now()}", email: "a-${Date.now()}@x.com", password: "testuser" }) { jwt } }`,
7  )
8)?.data?.register?.jwt;
9
10const testuser2Jwt = (
11  await gql(
12    `mutation { register(input: { username: "testuser2-${Date.now()}", email: "b-${Date.now()}@x.com", password: "testuser2" }) { jwt } }`,
13  )
14)?.data?.register?.jwt;
15
16const testuserNote = (
17  await gql(
18    `mutation { createNote(data: { title: "testuser-only", content: "secret" }) { documentId } }`,
19    undefined,
20    { Authorization: `Bearer ${testuserJwt}` },
21  )
22)?.data?.createNote;
23
24const testuser2View = await gql(`{ notes { documentId } }`, undefined, {
25  Authorization: `Bearer ${testuser2Jwt}`,
26});
27check(
28  "testuser2 does not see testuser's notes in the list",
29  !testuser2View?.data?.notes?.some(
30    (n) => n.documentId === testuserNote.documentId,
31  ),
32);
33
34const testuser2Attack = await gql(
35  `mutation A($id: ID!) { togglePin(documentId: $id) { documentId } }`,
36  { id: testuserNote.documentId },
37  { Authorization: `Bearer ${testuser2Jwt}` },
38);
39check(
40  "testuser2 cannot toggle pin on testuser's note (Policy Failed)",
41  testuser2Attack?.errors?.some((e) =>
42    /Policy Failed|Forbidden/i.test(e.message),
43  ),
44);

Run it:

npm run test:backend

Heads-up: the existing checks now need authentication. Sections 1 through 8 of test-graphql.mjs make anonymous GraphQL calls (no Authorization header). After Step 1.2 disabled anonymous reads on Note, the script bails at Section 1 with No active notes in the database and never reaches the new ownership block. Two ways to fix the script:

1. Register a user at the top and authenticate every request by passing { Authorization: \Bearer ${jwt}` }as the third argument to everygql(...)` call.
2. Skip Sections 1 through 8 if you only want to run the ownership block; comment out everything before // 9. Ownership: two-user isolation.

The two new ownership checks themselves work correctly — they register fresh users and pass their own JWTs — so once you have npm run test:backend running again, those two lines will be green.

Step 8: Wire up the frontend

Backend done. The frontend needs three things: pages to register and log in, a way to attach the JWT to every GraphQL call, and a route gate that sends signed-out users to /login.

8.1 Cookie helpers

Create lib/auth.ts:

1// starter-template/lib/auth.ts
2import { cookies } from "next/headers";
3
4export const JWT_COOKIE = "strapi_jwt";
5
6export async function getJwt(): Promise<string | undefined> {
7  const store = await cookies();
8  return store.get(JWT_COOKIE)?.value;
9}
10
11export async function setJwt(jwt: string): Promise<void> {
12  const store = await cookies();
13  store.set(JWT_COOKIE, jwt, {
14    httpOnly: true,
15    sameSite: "lax",
16    secure: process.env.NODE_ENV === "production",
17    path: "/",
18    maxAge: 60 * 60 * 24 * 30, // 30 days
19  });
20}
21
22export async function clearJwt(): Promise<void> {
23  const store = await cookies();
24  store.delete(JWT_COOKIE);
25}

httpOnly matters. The browser keeps the cookie but JavaScript cannot read it, so an XSS payload cannot steal the token. The cookie is only read on the server, by Server Components and Server Actions through cookies().

8.2 Attach the JWT in Apollo Client

Open lib/apollo-client.ts and add a fetch wrapper that adds the token when one is present:

1// starter-template/lib/apollo-client.ts
2import { HttpLink } from "@apollo/client";
3import {
4  registerApolloClient,
5  ApolloClient,
6  InMemoryCache,
7} from "@apollo/client-integration-nextjs";
8import { getJwt } from "./auth";
9
10const STRAPI_GRAPHQL_URL =
11  process.env.STRAPI_GRAPHQL_URL ?? "http://localhost:1337/graphql";
12
13export const { getClient, query, PreloadQuery } = registerApolloClient(() => {
14  return new ApolloClient({
15    cache: new InMemoryCache({
16      typePolicies: {
17        Note: { keyFields: ["documentId"] },
18        Tag: { keyFields: ["documentId"] },
19        UsersPermissionsUser: { keyFields: ["id"] },
20      },
21    }),
22    link: new HttpLink({
23      uri: STRAPI_GRAPHQL_URL,
24      fetchOptions: { cache: "no-store" },
25      fetch: async (url, init = {}) => {
26        const jwt = await getJwt();
27        const headers = new Headers(init.headers);
28        if (jwt) headers.set("Authorization", `Bearer ${jwt}`);
29        return fetch(url, { ...init, headers });
30      },
31    }),
32  });
33});

registerApolloClient runs the factory once per request, so getJwt() runs fresh for every render and Server Action. The token never reaches the browser bundle.

8.3 Login, register, me operations

Append to lib/graphql.ts:

1export const LOGIN = gql`
2  mutation Login($input: UsersPermissionsLoginInput!) {
3    login(input: $input) {
4      jwt
5      user {
6        id
7        username
8        email
9      }
10    }
11  }
12`;
13
14export const REGISTER = gql`
15  mutation Register($input: UsersPermissionsRegisterInput!) {
16    register(input: $input) {
17      jwt
18      user {
19        id
20        username
21        email
22      }
23    }
24  }
25`;
26
27export const ME = gql`
28  query Me {
29    me {
30      id
31      username
32      email
33    }
34  }
35`;

8.4 The login page

The starter already ships app/login/page.tsx with the form, the error banner, and a stub loginAction that just console.logs. Open the file if you want to see the JSX. To wire it up, replace the contents of app/login/actions.ts:

1// starter-template/app/login/actions.ts
2"use server";
3
4import { redirect } from "next/navigation";
5import { getClient } from "@/lib/apollo-client";
6import { LOGIN } from "@/lib/graphql";
7import { setJwt } from "@/lib/auth";
8
9const asString = (v: FormDataEntryValue | null) =>
10  typeof v === "string" ? v : "";
11
12export async function loginAction(formData: FormData) {
13  const identifier = asString(formData.get("identifier")).trim();
14  const password = asString(formData.get("password"));
15
16  if (!identifier || !password) return;
17
18  let jwt: string | undefined;
19  try {
20    const { data } = await getClient().mutate<{
21      login: { jwt: string };
22    }>({
23      mutation: LOGIN,
24      variables: { input: { identifier, password } },
25    });
26    jwt = data?.login?.jwt;
27  } catch {
28    // Apollo throws CombinedGraphQLErrors when Strapi returns an error
29    // (e.g. "Invalid identifier or password"). Catch it and fall through
30    // to the !jwt redirect below; otherwise the throw bubbles to Next's
31    // runtime overlay instead of giving the user an error message.
32  }
33
34  if (!jwt) redirect("/login?error=invalid");
35
36  await setJwt(jwt);
37  redirect("/");
38}