Wagtail MCP

🚧🤖 this is an early vibe-coded prototype of an MCP server for Wagtail. Proceed with caution! Until this is ready, you might prefer to use the wagtail-cli. See CMS with AI, not AI CMS: Wagtail 8.0’s new API for context.

A Model Context Protocol (MCP) server for Wagtail, exposing the Wagtail v3 API operations as model-callable tools.

Capabilities

• 60 curated tools covering the full shipped v3 operation surface: pages (incl. revisions, move/copy/revert, aliases, translations), images, documents, snippets, redirects, sites and locales — plus api_call / api_schema, an OpenAPI-backed escape hatch for anything else.
• In-process dispatch — tools call the mounted v3 API inside the process via Django's test client, so auth, permissions, validation and action logic are the real thing. No network hop, no half-reimplemented client.
• Bearer-token auth with Wagtail APIToken; the v3 API stays the single authority for authorization.
• Stateless Streamable HTTP — a Django view that works under WSGI and ASGI, with no extra processes.

Quickstart

wagtail-mcp is a Django app: add wagtail.api.v3 and wagtail_mcp, mount /api/v3/ and /mcp/, create an API token, point a client at /mcp/. See the full getting started.

Point a remote MCP client at the endpoint:

{
  "mcp": {
    "wagtail": {
      "type": "remote",
      "url": "http://localhost:8000/mcp/",
      "headers": {
        "Authorization": "Bearer <your-api-token>",
        "Content-Type": "application/json"
      }
    }
  }
}

Try the demo

The repo ships demo/, a Wagtail site that mounts both the API and the MCP endpoint and seeds a demo token:

just demo

Then use demo/'s token (see demo/README.md) in your client config. The editorial scenario test is a complete worked example an agent can follow.

Documentation

Using the package: docs.

• Getting started
• Tool reference
• Configuration
• Admin agent
• Escape hatch
• Limitations

Changing the package: contributing.

Supported versions

This package supports Wagtail 8.0 and up (the v3 API preview), Django 5.2/6.0, and the compatible Python versions.

Development

Install dependencies, run the tests, lint, and run the demo (asgiref.sync handles the transport, so just test and just lint are the normal gates):

just install
just test
just lint
just demo

See CONTRIBUTING.md for setup, tests, and releases.