canlii-mcp

An MCP (Model Context Protocol) server for the CanLII Canadian legal information API. Gives AI assistants access to Canadian case law and legislation metadata across all federal, provincial, and territorial jurisdictions.

Forked from tomilashy/canlii-mcp. This fork adds bring-your-own-key (BYOK) auth, a /health route, and a hosted endpoint at canlii-mcp.vaquill.ai. Tools are unchanged.

Note: The CanLII API provides metadata only — titles, citations, dates, keywords, and citation relationships. Full document text is not available through the API.

Use the hosted endpoint (no install)

https://canlii-mcp.vaquill.ai/mcp

Two headers are required for the hosted instance:

• Authorization: Bearer <MCP_AUTH_TOKEN> — gates access to the MCP server itself
• X-CanLII-Token: <your_canlii_api_key> — your CanLII key. Apply at canlii.org/en/api/. The server never stores your key.

Claude Desktop / Claude Code

{
  "mcpServers": {
    "canlii": {
      "url": "https://canlii-mcp.vaquill.ai/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_MCP_TOKEN",
        "X-CanLII-Token": "YOUR_CANLII_API_KEY"
      }
    }
  }
}

Cursor / VS Code / Windsurf

Same pattern: any client supporting MCP streamable HTTP with custom headers works. For stdio-only clients use mcp-remote to proxy.

Authentication

Tools

Requirements

• Node.js 22+
• A CanLII API key — apply here

Usage

stdio via npx (quickest)

{
  "mcpServers": {
    "canlii": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@tomilashy/canlii-mcp"],
      "env": {
        "CANLII_API": "your_api_key"
      }
    }
  }
}

stdio (from source)

npm install
npm run build
node dist/index.js

Add to your MCP config:

{
  "mcpServers": {
    "canlii": {
      "command": "node",
      "args": ["/path/to/canlii-mcp/dist/index.js"],
      "env": {
        "CANLII_API": "your_api_key"
      }
    }
  }
}

HTTP server

PORT=3000 CANLII_API=your_api_key node dist/index.js --transport http

The MCP endpoint is available at http://localhost:3000/mcp. The server runs in stateless mode — each request is self-contained, no session ID or initialize handshake required. Clients can call tools directly:

curl -X POST http://localhost:3000/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_case_databases","arguments":{"language":"en"}}}'

Docker

docker run -e CANLII_API=your_api_key -e MCP_AUTH_TOKEN=your_secret -p 3000:3000 ghcr.io/tomilashy/canlii-mcp

Or with Docker Compose:

services:
  canlii-mcp:
    image: ghcr.io/tomilashy/canlii-mcp
    environment:
      CANLII_API: your_api_key
      MCP_AUTH_TOKEN: your_secret  # optional
    ports:
      - "3000:3000"

Cloudflare Workers

The server includes a Workers-compatible entry point (src/worker.ts).

CLI deploy

npx wrangler secret put CANLII_API
npx wrangler secret put MCP_AUTH_TOKEN  # optional
npx wrangler deploy

Dashboard deploy (Connect to Git)

1. Go to Cloudflare Dashboard → Workers & Pages → Create → Connect to Git
2. Select your tomilashy/canlii-mcp repository
3. On the Set up your application page:
   • Project name: canlii-mcp
   • Build command: npm install && npm run build
   • Deploy command: npx wrangler deploy (pre-filled)
4. Expand Advanced settings:
   • Variable name: CANLII_API
   • Variable value: your CanLII API key
   • Check Encrypt to store it as a secret
5. Click Deploy

The MCP endpoint will be at https://canlii-mcp.<your-subdomain>.workers.dev/mcp.

Configuration

Rate Limits

The server enforces CanLII's API limits automatically:

• 1 request at a time
• 2 requests per second
• 5,000 requests per day

Requests that exceed the daily limit return an error rather than hitting the API.

Development

npm install
npm run build      # compile TypeScript
npm run watch      # watch mode

Release

This project uses Semantic Versioning via semantic-release. Commit messages follow the Conventional Commits spec:

Pushing to main triggers the release workflow. If a release is cut, the Docker image is automatically built and published to ghcr.io.

License

MIT