Fourth Playwright MCP Server

Remote MCP server combining FastMCP v3 (Python) orchestration with @playwright/mcp (Node.js) browser automation. Designed for deployment to Railway as a Claude.ai Custom Connector.

Architecture

FastMCP v3 (Python) ─── SSE/Streamable HTTP ──→ Claude.ai
       │
       └── subprocess (stdio) ──→ @playwright/mcp (Node.js)

• FastMCP v3: OAuth 2.1 proxy, Skills system, custom tools, HTTP transport

• @playwright/mcp: 70+ browser automation tools via stdio subprocess

• Custom tools: Fourth-specific authentication, navigation, data extraction

Setup

Prerequisites

• Python 3.11+

• Node.js 18+

• uv (Python package manager)

Install

# Python dependencies
uv sync

# Node dependencies (installs @playwright/mcp + Chromium)
npm install

Run locally

uv run python src/server.py
# Server starts on http://localhost:8000
# Health check: curl http://localhost:8000/health
# MCP endpoint: http://localhost:8000/mcp

List tools

uv run fastmcp list src/server.py

Deployment (Railway)

1. Connect this repo to Railway

2. Railway auto-detects the Dockerfile

3. Set environment variables (see .env.example)

4. Railway sets PORT automatically

Claude.ai Custom Connector

1. Go to Claude.ai → Settings → Connectors → Add Custom

2. SSE URL: https://<app>.railway.app/mcp

3. Configure OAuth with your Client ID/Secret

Custom Fourth Tools

Skills

Skills are exposed as MCP resources from the skills/ directory:

• browser-automation - General browser automation patterns

• fourth-workflows - Fourth-specific workflow instructions

Environment Variables

See .env.example for all configuration options.