MCP for Flarum
# MCP for Flarum
[](https://github.com/linkrobins/mcp-for-flarum/actions/workflows/ci.yml)
[](LICENSE)
A [Model Context Protocol](https://modelcontextprotocol.io) server for [Flarum](https://flarum.org).
It gives any MCP-compatible AI client (Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Zed, and others) full access to a Flarum forum's API: read and search discussions and posts, create threads and replies, manage users, tags and groups, moderate content, change settings, and call any third-party extension endpoint.
> Flarum's whole API is uniform [JSON:API](https://jsonapi.org), so a small set of generic tools covers the **entire** surface, including extensions, rather than hundreds of hand-written ones.
## Tools
**Generic (full API coverage):**
| Tool | What it does |
| --- | --- |
| `flarum_list` | List/search any resource type with filters, includes, sort, pagination |
| `flarum_get` | Fetch one resource by type and id |
| `flarum_create` | Create any resource |
| `flarum_update` | Update any resource (also lock/sticky/approve for moderation) |
| `flarum_delete` | Delete any resource |
| `flarum_request` | Raw escape hatch for any endpoint |
**Convenience:**
| Tool | What it does |
| --- | --- |
| `flarum_whoami` | Forum info + the user the API key acts as |
| `flarum_search` | Full-text discussion search |
| `flarum_create_discussion` | Start a thread (title + content + optional tags) |
| `flarum_reply` | Reply to a discussion |
**Official docs (on by default):**
Search and read Flarum's official [2.0 documentation](https://docs.flarum.org/2.x/) so the AI can check how a setting, permission, extender, or REST endpoint is meant to work before acting. These read the public docs only (never your forum or its key), so they work in any mode, including read-only and with no API key. They read the live docs, so results always reflect the current 2.0 documentation. Turn them off with `FLARUM_DOCS=0`.
| Tool | What it does |
| --- | --- |
| `flarum_docs_search` | Search the 2.0 docs; returns ranked pages with snippets |
| `flarum_docs_get` | Read a full docs page as Markdown (by slug, path, or URL) |
| `flarum_docs_list` | List/browse the available 2.0 docs pages |
**Extension development (on by default):**
A development reference for building or reviewing a Flarum 2.0 extension: scaffolding and architecture, `composer.json`, the TypeScript frontend, backend (API resources/models/migrations), i18n, testing, static analysis & CI, and releasing. Combines the conventions the official docs establish, the de-facto FriendsOfFlarum standard, and patterns that prevent real production bugs. Static content, so it works in any mode, including read-only and with no API key. Turn it off with `FLARUM_DEV=0`.
| Tool | What it does |
| --- | --- |
| `flarum_dev` | Returns the extension-development reference; optional `topic` (scaffold, composer, frontend, backend, i18n, testing, quality-ci, release) to narrow it |
**Troubleshooting (on by default):**
Plain-language help for a broken or misbehaving forum, aimed at admins rather than developers: safe first-aid fixes, how to run `php flarum info` and find your logs on typical hosting, what the common errors actually mean, and how to write a redacted support request. Static knowledge that needs no server access, so it works in any mode, including read-only and with no API key. Turn it off with `FLARUM_TROUBLESHOOT=0`.
| Tool | What it does |
| --- | --- |
| `flarum_troubleshoot` | Returns the troubleshooting guide; optional `topic` (first-aid, info, logs, common, report) to narrow it |
**Extension management (opt-in, off by default):**
Registered only when `FLARUM_EXTENSIONS=1` and the forum has the official [`flarum/extension-manager`](https://github.com/flarum/extension-manager) installed. These drive Composer on the server, so they need an admin key and write mode. See [Managing extensions](#managing-extensions).
| Tool | What it does |
| --- | --- |
| `flarum_ext_search` | Search Packagist for installable extensions, language packs, or themes |
| `flarum_ext_why_not` | Dry-run compatibility check for a package (changes nothing) |
| `flarum_ext_install` | Install an extension (optionally enable it after) |
| `flarum_ext_update` | Update one extension (`soft` in-range, or `hard` to latest) |
| `flarum_ext_remove` | Uninstall an extension |
| `flarum_ext_toggle` | Enable/disable an installed extension (no Composer; instant) |
| `flarum_ext_check_updates` | Check Packagist for available updates |
| `flarum_ext_bulk_update` | Bulk update: `minor`, `major` (with `dryRun`), or `global` |
| `flarum_ext_configure_composer` | Read/set `minimum-stability`, repositories, or private registry auth |
| `flarum_ext_tasks` | List install/update job history and Composer output (poll async jobs) |
**Control-plane hooks (dormant unless you wire them up):**
Two capabilities stay inactive unless you point them at a service that answers them. They are gated on `DIAG_URL` and `SNAPSHOT_URL`; with those unset, which is the default, the diagnostic tools never register and the snapshot hook is a no-op. Nothing here is closed off — the code is [MIT](#license) like the rest, so you can point them at a backend of your own.
| Capability | Gate | What it does |
| --- | --- | --- |
| `flarum_diag` / `flarum_triage` | `DIAG_URL` + token | Read-only troubleshooting through a hosting control plane (boot errors, post-update breakage, mail/queue failures), which keeps working even when Flarum itself won't boot. See [docs/managed-troubleshooting.md](docs/managed-troubleshooting.md). |
| Pre-change snapshot | `SNAPSHOT_URL` + token | Best-effort restore point taken before the first write of a session. See `SNAPSHOT_URL` under [Hosting](#hosting-http-transport). |
Every other tool above (generic, convenience, docs, dev, and extension management) works out of the box with no such backend.
## Prompts
The server also ships prompts: named, ready-made workflows that chain the right tools in the right order, so you don't have to describe the whole job yourself. Clients surface them in their own way (Claude Code lists them as slash commands, Claude Desktop under the prompts menu).
| Prompt | What it does |
| --- | --- |
| `build-flarum-extension` | Scaffolds and builds a Flarum 2.0 extension, following the `flarum_dev` contracts |
| `review-flarum-extension` | Reviews existing extension code against those same contracts |
| `check-flarum-compatibility` | Checks an extension against real production stacks (queue drivers, Redis, multi-server, sub-path URLs) |
| `prepare-flarum-support-request` | Assembles a redacted support request from a forum that still loads |
The first three come with the extension-development reference (`FLARUM_DEV`), the last with the troubleshooting guide (`FLARUM_TROUBLESHOOT`); disabling either hides its prompts too.
## Configuration
| Variable | Required | Description |
| --- | --- | --- |
| `FLARUM_URL` | yes | Your forum's base URL, e.g. `https://discuss.example.com` |
| `FLARUM_API_KEY` | for writes / private data | A Flarum API key (from the `api_keys` table). Without it, only public read access is available. |
| `FLARUM_USER_ID` | optional | Act as this user id when using a master API key |
| `FLARUM_MODE` | optional | `write` (default) or `read`. In `read` mode the server refuses every mutating request (create/update/delete and any non-GET `flarum_request`) and the write tools are hidden. `READ_ONLY=1` does the same. Use it to point an AI at a real forum without risking changes. |
| `FLARUM_EXTENSIONS` | optional | `1`/`true` registers the extension-management tools (install/update/remove extensions via `flarum/extension-manager`). Off by default; requires write mode and an admin key. See [Managing extensions](#managing-extensions). |
| `FLARUM_DOCS` | optional | On by default. Set `0`/`false`/`off` to hide the official-docs tools (`flarum_docs_search`/`get`/`list`). They read the public docs only, never your forum. |
| `FLARUM_DEV` | optional | On by default. Set `0`/`false`/`off` to hide the extension-development reference tool (`flarum_dev`) and its prompts. Static guidance; never touches your forum. |
| `FLARUM_TROUBLESHOOT` | optional | On by default. Set `0`/`false`/`off` to hide the troubleshooting guide (`flarum_troubleshoot`) and its prompt. Static guidance; never touches your forum. |
| `FLARUM_TIMEOUT` | optional | Request timeout in ms (default 30000) |
| `FLARUM_USER_AGENT` | optional | Override the `User-Agent` sent to your forum. Defaults to `mcp-for-flarum/<version> (+repo url)`. See [Behind Cloudflare or a WAF](#behind-cloudflare-or-a-waf). |
### Behind Cloudflare or a WAF
Many Flarum forums sit behind Cloudflare. Some WAF configurations block requests whose `User-Agent` looks scripted or empty, returning **Cloudflare error 1010** (`browser_signature_banned`) before the request ever reaches Flarum. The server sends a descriptive, identifiable User-Agent by default for exactly this reason, so the common case works out of the box.
If your forum still blocks it, allowlist the tool rather than loosening your firewall:
- **Allowlist the User-Agent.** In Cloudflare, add a WAF rule like `User-Agent contains "mcp-for-flarum"` → *Skip* / *Allow*. The default UA is `mcp-for-flarum/<version> (+https://github.com/linkrobins/mcp-for-flarum)`.
- **Or allowlist the server IP** (best for a hosted/single-source deployment).
- **Or set a custom UA** with `FLARUM_USER_AGENT` to match an existing allow rule.
Do not work around this by spoofing a browser User-Agent: it is fragile and makes the traffic impossible to allowlist or audit.
### Getting an API key
Flarum has no admin UI for API keys yet. Create one directly in the database:
```sql
INSERT INTO api_keys (`key`, user_id, created_at)
VALUES (REPLACE(UUID(), '-', ''), 1, NOW());
```
Use the resulting `key` as `FLARUM_API_KEY`. Setting `user_id` (or `FLARUM_USER_ID`) makes the key act as that user, so its permissions are exactly that user's permissions.
### Managing extensions
Set `FLARUM_EXTENSIONS=1` to let the AI install, update, remove, enable, and disable extensions. This is **off by default** because it runs Composer on your server and can change what code your forum runs, which is far more powerful than editing content. It requires all of:
- The official [`flarum/extension-manager`](https://github.com/flarum/extension-manager) installed and enabled on the forum.
- Write mode (not `FLARUM_MODE=read`) and an API key whose user is an **admin**.
- A server that can actually run Composer: the PHP functions `proc_open` and `escapeshellarg` available, and `vendor/`, `storage/`, `composer.json`, and `composer.lock` writable.
How long-running installs are reported depends on your forum's queue:
- **Background queue** (Redis, database, etc. with a running worker): the call returns once the job finishes. The tools poll the manager's task list for you and return the Composer output. If no worker is consuming jobs, the call times out and says so rather than hanging.
- **Synchronous** (`sync` queue, or the manager's "run jobs in background" setting off): the request blocks until Composer finishes and returns the result inline. Very large updates can hit PHP/gateway timeouts even though Composer keeps running.
`flarum_ext_install` does not enable the extension unless you pass `enable: true`. Use `flarum_ext_why_not` first to confirm a package is compatible with your Flarum version. For bulk or major updates, take a backup first.
**Enabling can break a forum.** The manager refuses to install an extension whose published Flarum compatibility doesn't match your version, but that check is best-effort: it reads the latest stable release's declared `flarum/core` constraint, is skipped when Packagist is unreachable or the package declares nothing, and the version Composer actually installs can differ from the one it checked. So an extension can pass the check and still fail to boot when enabled, taking the whole site down (every page, including the admin panel and this tool's own API, starts returning a 500). The manager cannot then disable it for you: recovery means removing the extension from the `extensions_enabled` setting in the database and `composer remove`-ing it by hand. This is why install does not auto-enable by default, and why a backup before enabling unfamiliar extensions is worth it.
## Install & run
Two one-line options, neither of which needs a checkout: a Docker image, or `npx` pointed straight at this repository. Pick whichever runtime you already have.
Both accept the same configuration. Leave out `FLARUM_API_KEY` to start with public read-only access, or set `FLARUM_MODE=read` to guarantee the AI can never change anything.
### Option 1: Docker
For a local Claude client (stdio):
```bash
docker run -i --rm \
-e FLARUM_URL=https://discuss.example.com \
-e FLARUM_API_KEY=xxxxx \
ghcr.io/linkrobins/mcp-for-flarum
```
Claude Code:
```bash
claude mcp add flarum -- docker run -i --rm -e FLARUM_URL=https://discuss.example.com -e FLARUM_API_KEY=xxxxx ghcr.io/linkrobins/mcp-for-flarum
```
Claude Desktop / Cursor / Windsurf (JSON config):
```json
{
"mcpServers": {
"flarum": {
"command": "docker",
"args": ["run", "-i", "--rm", "-e", "FLARUM_URL", "-e", "FLARUM_API_KEY", "ghcr.io/linkrobins/mcp-for-flarum"],
"env": {
"FLARUM_URL": "https://discuss.example.com",
"FLARUM_API_KEY": "xxxxx"
}
}
}
}
```
### Option 2: npx, straight from this repo (no Docker)
If you have [Node.js](https://nodejs.org) 18+ and git, `npx` can fetch, build, and run it in one step. There is nothing to clone and nothing to install.
Claude Code:
```bash
claude mcp add flarum \
-e FLARUM_URL=https://discuss.example.com \
-e FLARUM_API_KEY=xxxxx \
-- npx -y github:linkrobins/mcp-for-flarum
```
Claude Desktop / Cursor / Windsurf / VS Code / Zed (JSON config):
```json
{
"mcpServers": {
"flarum": {
"command": "npx",
"args": ["-y", "github:linkrobins/mcp-for-flarum"],
"env": {
"FLARUM_URL": "https://discuss.example.com",
"FLARUM_API_KEY": "xxxxx"
}
}
}
}
```
The first run compiles the TypeScript, so it takes a few seconds; npx caches the result and later starts are fast. Pin a release by appending a tag, e.g. `github:linkrobins/mcp-for-flarum#v0.8.0`, which is worth doing for anything long-lived. To install it once instead of resolving on each start:
```bash
npm install -g github:linkrobins/mcp-for-flarum
```
and use `mcp-for-flarum` as the `command`.
> This project is not published to the npm registry, so `npm install mcp-for-flarum` will not find it (and if that name ever does resolve, it is not this project). Use the `github:` form above.
### Option 3: From source
```bash
git clone https://github.com/linkrobins/mcp-for-flarum.git
cd mcp-for-flarum
npm install && npm run build
FLARUM_URL=https://discuss.example.com FLARUM_API_KEY=xxxxx node dist/index.js
```
Then point your client's `command` at `node /absolute/path/to/mcp-for-flarum/dist/index.js`.
## Hosting (HTTP transport)
The same binary can run as a long-lived web service over **Streamable HTTP**, so you can host it instead of running it locally. This is what web-based clients (which can't spawn a local process) connect to.
Start in HTTP mode with `--http` (or `MCP_TRANSPORT=http`):
```bash
FLARUM_URL=https://discuss.example.com \
FLARUM_API_KEY=xxxxx \
MCP_AUTH_TOKEN=a-long-random-secret \
PORT=3000 \
npx -y github:linkrobins/mcp-for-flarum --http
```
(From a source checkout, `node dist/index.js --http` does the same.)
It serves:
- `POST /mcp`, the MCP endpoint (Streamable HTTP, stateless)
- `GET /health`, health check for load balancers / uptime monitors
Hosting-specific configuration:
| Variable | Default | Description |
| --- | --- | --- |
| `MCP_TRANSPORT` | `stdio` | Set to `http` to run the web service (or pass `--http`) |
| `PORT` | `3000` | HTTP port |
| `HOST` | `127.0.0.1` | Bind address. Fails closed: it **refuses to bind a non-localhost address unless `MCP_AUTH_TOKEN` is set**. To expose it (e.g. in Docker), set `HOST=0.0.0.0` and a token. |
| `MCP_AUTH_TOKEN` | _(none)_ | If set, requests must send `Authorization: Bearer <token>`. Required to expose a non-localhost interface. |
| `SNAPSHOT_URL` | _(none)_ | Optional control-plane hook. When set, the server fires a best-effort `POST` here before the **first write** in a session, so a restore point can be taken before AI-driven edits. Failures never block writes; leave it unset unless you run a backend that answers it. |
| `SNAPSHOT_TOKEN` | `MCP_AUTH_TOKEN` | Bearer token sent with the `SNAPSHOT_URL` ping. Defaults to `MCP_AUTH_TOKEN`. |
> Security: a hosted instance can read, write, and delete on the forum its key targets. Always run it behind TLS, set `MCP_AUTH_TOKEN` (or front it with your own auth/OAuth proxy), and give the API key's user the least privilege it needs. The Flarum API key stays server-side and is never exposed to clients.
### Docker (hosted mode)
```bash
docker run -p 3000:3000 \
-e MCP_TRANSPORT=http \
-e HOST=0.0.0.0 \
-e FLARUM_URL=https://discuss.example.com \
-e FLARUM_API_KEY=xxxxx \
-e MCP_AUTH_TOKEN=a-long-random-secret \
ghcr.io/linkrobins/mcp-for-flarum
```
`HOST=0.0.0.0` is needed so the published port is reachable; the server only allows it because `MCP_AUTH_TOKEN` is set.
Or use the included [`docker-compose.yml`](docker-compose.yml): set your values and `docker compose up -d`.
## Development
```bash
npm install
npm run build
FLARUM_URL=... FLARUM_API_KEY=... node dist/index.js
```
## License
[MIT](LICENSE) © Link Robins. Free and open source — self-host it, modify it, and use it however you like.
**Commercial and hosted use is welcome, and needs no permission.** Run it for clients, bundle it into a paid product, offer it as a hosted service, fork it, rebrand it: MIT already allows all of that, and there is no separate license to buy, no key to obtain, and no tier that unlocks anything. Nothing in the code is held back — the two control-plane hooks above are dormant integration points, not a paywall, and you are free to point them at infrastructure of your own. The only ask is the one the license makes: keep the copyright notice. If you build something with it, I would love to hear about it.
## Trademarks
Flarum is a trademark of the Flarum Foundation. This is an independent project that works *with* Flarum via its API; it is not affiliated with, endorsed by, or sponsored by the Flarum Foundation.
TDQS
Scored across 15 tools
Most tools have clear roles, and the generic CRUD verbs (flarum_create/update/delete) are easy to distinguish from the docs and reference tools. However, flarum_list vs flarum_search and flarum_create vs flarum_create_discussion overlap in behavior, even though the descriptions point to the specialized helper being more convenient.
All tools share the flarum_ prefix and use readable verbs, but the conventions are mixed: core API verbs are single-word (flarum_get, flarum_update), docs tools reverse the order (flarum_docs_get, flarum_docs_search), and flarum_create_discussion follows verb_noun. flarum_whoami, flarum_dev, and flarum_troubleshoot are standalone names that don't fit a verb_noun pattern.
Fifteen tools is within the ideal range and matches a broad purpose: Flarum API CRUD, discussion helpers, documentation lookup, and admin/dev guidance. The generic resource tools plus convenience wrappers and an escape hatch give the set a clear rationale without feeling bloated.
The generic flarum_list/get/create/update/delete tools provide near-complete lifecycle coverage for any Flarum resource, and flarum_request covers any remaining API endpoint. Discussion creation/reply, docs search/get, and whoami cover the main forum-administration and learning workflows, leaving no obvious dead ends.