DocketBird MCP Server

An MCP server for searching and downloading court documents via the DocketBird API. Deployed on DigitalOcean with Docker, using OAuth 2.0 so each user brings their own DocketBird API key.

Tools

Research tools work across DocketBird's full index (all courts, all cases); account tools act on your firm's own DocketBird data.

Using these tools from an agent: a ready-to-install Claude skill lives in skills/docketbird-mcp/, covering the case-ID format, each tool, and common research workflows.

Related MCP server: legal-mcp

Requirements

• Python 3.11

• uv package manager

Setup

1. Install uv:

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Create and activate a virtual environment:

uv venv
source .venv/bin/activate

3. Install dependencies:

uv pip install -r requirements.txt

Running the Server

# stdio transport (uses DOCKETBIRD_API_KEY env var, no OAuth)
DOCKETBIRD_API_KEY="your-key" python docketbird_mcp.py --transport stdio

# HTTP transport with OAuth (Streamable HTTP at /mcp)
python docketbird_mcp.py --transport http
# Then visit http://localhost:8080/signup to create an account

Note: All diagnostic logging goes to stderr, never stdout. This keeps the stdio JSON-RPC stream clean — writing logs to stdout would corrupt it and break the client.

Environment Variables

See .env.example for a template.

Connecting to the Deployed Server

See docs/CONNECTING.md for the full walkthrough, or DocketBird_MCP_Installation_Guide.pdf for the same guide with screenshots.

Quick version

1. Register at https://app.docketbird-mcp.com/signup with your email, password, and DocketBird API key

2. In Claude.ai or Claude Desktop, add a remote MCP server with URL https://app.docketbird-mcp.com/mcp

3. Claude auto-discovers OAuth, redirects you to log in, and connects

Stdio (local development)

For Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json) or Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "docketbird-mcp": {
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "/path/to/docketbird-mcp-plus",
        "python",
        "docketbird_mcp.py"
      ],
      "env": {
        "DOCKETBIRD_API_KEY": "YOUR_KEY"
      }
    }
  }
}

Authentication

The server uses OAuth 2.0 with PKCE for HTTP mode. Each user registers with their own DocketBird API key, which is stored server-side and attached to OAuth tokens. The SDK handles the protocol endpoints automatically:

• /.well-known/oauth-authorization-server - OAuth metadata discovery

• /register - Dynamic Client Registration

• /authorize - Authorization endpoint (redirects to /login)

• /token - Token exchange and refresh

In stdio mode, the DOCKETBIRD_API_KEY env var is used directly (no OAuth).

Security

• OAuth 2.0 with PKCE (no shared API key on the server)

• Per-user DocketBird API keys stored in SQLite with bcrypt-hashed passwords

• Rate limiting: 30 requests per 60 seconds per IP

• HTTPS-only downloads with SSRF domain allowlist

• Path traversal protection on file downloads

• Container runs as non-root mcpuser

• GitHub Actions pinned to commit SHAs

• Dependencies pinned to exact versions

• Expired tokens, auth codes, and pending sessions are purged hourly (in both stdio and HTTP modes)

Downloads — where files go: over a remote (HTTP) connection the download tools return document content and links to your client; a save_path is ignored (it would write to the server, not your machine). In local stdio mode, pass a save_path to save to your own machine.

Development & Testing

Install the dev extras and run the test suite:

uv pip install -e ".[dev]"   # or: pip install -e ".[dev]"
pytest

The suite (tests/) runs fully offline — network calls are faked — and covers the security helpers (path/URL validation, filename sanitization), pagination math, the rate limiter, error formatting, the streaming download size cap, and the remote-vs-local download behavior (inline content / links vs save-to-disk).

For a deeper look at how the server is wired together (OAuth flow, request lifecycle, database schema, security model), see docs/ARCHITECTURE.md.

Troubleshooting

• stdio client shows JSON parse errors: ensure nothing in your environment writes to stdout. This server logs to stderr by design; custom forks should keep it that way.

• No DocketBird API key available: in stdio mode set DOCKETBIRD_API_KEY; in HTTP mode complete the OAuth login at /signup → connect from Claude.

• OAuth login loops or "Invalid redirect URI": confirm SERVER_URL exactly matches the public URL clients use (scheme + host, no trailing slash).

• Changed your DocketBird API key but tools still fail: use /change-api-key. It validates the new key against DocketBird and clears stale access tokens so the new key takes effect immediately.

Deployment

Deployed via Docker and GitHub Actions. Pushes to main trigger automatic deployment.

• Domain: app.docketbird-mcp.com

• Docker volume: docketbird-data at /app/data (SQLite auth database)

• Health check: https://app.docketbird-mcp.com/health — returns {"status":"ok","service":"docketbird-mcp","version":"<git-sha>"}, where version is the deployed commit (set by the deploy workflow via GIT_SHA), so a single curl confirms exactly which build is live.

• Caddy reverse proxy handles HTTPS (Let's Encrypt)

Local Docker Build

docker build -t docketbird-mcp:latest .

docker run -d \
  --name docketbird-mcp \
  --restart=always \
  -e SERVER_URL="http://localhost:8040" \
  -v docketbird-data:/app/data \
  -p 8040:8080 \
  docketbird-mcp:latest

Reference Data

• case_types.json - Case type abbreviations and examples (no API endpoint exists for these)

• courts.json - Historical snapshot of court codes, frozen 2026-05-31. No longer read by any tool — docketbird_list_courts serves the live GET /courts endpoint.

Acknowledgment

This project is built upon the original docketbird-mcp developed in conjunction with @federicoburman and the Gravix.AI team.