LangChain Agent MCP Server

A production-ready MCP server exposing LangChain agent capabilities through the Model Context Protocol, deployed on Google Cloud Run.

🚀 Overview

This is a standalone backend service that wraps a LangChain agent as a single, high-level MCP Tool. The server is built with FastAPI and deployed on Google Cloud Run, providing a scalable, production-ready solution for exposing AI agent capabilities to any MCP-compliant client.

Live Service: https://langchain-agent-mcp-server-554655392699.us-central1.run.app

✨ Features

• ✅ MCP Compliance - Full Model Context Protocol support

• ✅ LangChain Agent - Multi-step reasoning with ReAct pattern

• ✅ Playwright Sandbox - Interactive preview of accessibility snapshots (NEW!)

• ✅ Google Cloud Run - Scalable, serverless deployment

• ✅ Tool Support - Extensible framework for custom tools

• ✅ Production Ready - Error handling, logging, and monitoring

• ✅ Docker Support - Containerized for easy deployment

🏗️ Architecture

🛠️ Quick Start

Prerequisites

• Python 3.11+

• OpenAI API key

• Google Cloud account (for Cloud Run deployment)

• Docker (optional, for local testing)

Local Development

1. Clone the repository:

   git clone https://github.com/mcpmessenger/LangchainMCP.git cd LangchainMCP

2. Install dependencies:

   # Windows py -m pip install -r requirements.txt # Linux/Mac pip install -r requirements.txt

3. Set up environment variables: Create a .env file:

   OPENAI_API_KEY=your-openai-api-key-here OPENAI_MODEL=gpt-4o-mini PORT=8000

4. Run the server:

   # Windows py run_server.py # Linux/Mac python run_server.py

5. Test the endpoints:

   • Health: http://localhost:8000/health

   • Manifest: http://localhost:8000/mcp/manifest

   • API Docs: http://localhost:8000/docs

   • Playwright Sandbox: http://localhost:8080/sandbox (after starting frontend)

6. Start the frontend (optional):

   # Install frontend dependencies (first time only) npm install # Start frontend dev server npm run dev

   Then visit http://localhost:8080/sandbox to use the Playwright Sandbox preview feature.

☁️ Google Cloud Run Deployment

The server is designed for deployment on Google Cloud Run. See our comprehensive deployment guides:

• DEPLOY_CLOUD_RUN_WINDOWS.md - Windows deployment guide

• DEPLOY_CLOUD_RUN.md - General deployment guide

• QUICK_DEPLOY.md - Quick reference

Quick Deploy

Current Deployment

• Service URL: https://langchain-agent-mcp-server-554655392699.us-central1.run.app

• Project: slashmcp

• Region: us-central1

• Status: ✅ Live and operational

📡 API Endpoints

MCP Endpoints

Get Manifest

Returns the MCP manifest declaring available tools.

Response:

Invoke Tool

With System Instruction (Optional):

Response:

System Instructions

The agent_executor tool supports an optional system_instruction parameter that allows you to customize the agent's behavior on a per-invocation basis.

Usage:

• Basic Query (uses default prompt):

  { "tool": "agent_executor", "arguments": { "query": "What is the weather today?" } }

• Query with Custom Instruction:

  { "tool": "agent_executor", "arguments": { "query": "Explain quantum computing", "system_instruction": "You are a physics professor. Explain concepts clearly and use examples." } }

• Personality Customization:

  { "tool": "agent_executor", "arguments": { "query": "Tell me about space", "system_instruction": "You are a pirate explaining complex topics. Use pirate terminology!" } }

Notes:

• If system_instruction is omitted, the agent uses its default prompt

• Empty or whitespace-only instructions are ignored (default prompt is used)

• Each invocation with a custom instruction creates a new agent instance

Playwright Sandbox Endpoints

Generate Accessibility Snapshot

Response:

Features:

• Generates structured accessibility snapshots of any website

• Shows how AI "views" websites through structured data

• Caching support for popular sites

• Token count estimation

• Windows-compatible (uses ProactorEventLoop)

Test Prompt Against Snapshot

Response:

Playwright Sandbox UI: Visit http://localhost:8080/sandbox to use the interactive preview feature:

• Enter any URL to generate a snapshot

• View live website side-by-side with AI accessibility snapshot

• Test prompts to find elements in the snapshot

• See token savings compared to full HTML/screenshots

Other Endpoints

• GET / - Server information

• GET /health - Health check

• GET /api/tasks - Safe task summaries (optional monitoring)

• GET /api/tasks/{task_id} - Safe task summary (optional monitoring)

• GET /docs - Interactive API documentation (Swagger UI)

🔧 Configuration

Environment Variables

📚 Documentation

📖 Full Documentation Site - Complete documentation with examples (GitHub Pages)

Quick Links:

• Getting Started - Set up and run locally

• Examples - Code examples including "Build a RAG agent in 10 lines"

• Deployment Guide - Deploy to Google Cloud Run

• API Reference - Complete API documentation

• Troubleshooting - Common issues and solutions

Build Docs Locally:

Additional Guides:

• README_BACKEND.md - Complete technical documentation

• PLAYWRIGHT_SANDBOX_SETUP.md - Playwright Sandbox setup and usage

• BUG_REPORT_PLAYWRIGHT_NOTIMPLEMENTEDERROR.md - Windows compatibility fix documentation

• DEPLOY_CLOUD_RUN_WINDOWS.md - Windows deployment guide

• INSTALL_PREREQUISITES.md - Prerequisites installation

• SLASHMCP_INTEGRATION.md - SlashMCP integration guide

• docs/glazyr-integration.md - Glazyr integration notes (screenshots → MCP invoke)

🧪 Testing

🎭 Playwright Sandbox Feature

The Playwright Sandbox is an interactive preview feature that demonstrates how AI agents "view" websites through structured accessibility data. This feature is particularly useful for understanding the value of structured snapshots compared to full HTML or screenshots.

Features

• Dual-View Interface: See the live website alongside its structured accessibility snapshot

• Token Efficiency: Compare token counts - snapshots are typically 90%+ smaller than full HTML

• Interactive Testing: Test prompts to find elements in the snapshot

• Caching: Popular sites are cached for faster demo results

• Windows Compatible: Fixed NotImplementedError on Windows using ProactorEventLoop

Quick Start

1. Install Playwright:

   py -m pip install playwright py -m playwright install chromium

2. Start Backend:

   py run_server.py

3. Start Frontend:

   npm install # First time only npm run dev

4. Visit Sandbox: Open http://localhost:8080/sandbox and try URLs like:

   • wikipedia.org

   • github.com

   • google.com

How It Works

1. Enter a URL - The system navigates to the website using Playwright

2. Generate Snapshot - Extracts structured accessibility information (roles, names, descriptions)

3. View Comparison - See the live site vs. the AI's structured view

4. Test Prompts - Try asking the AI to find specific elements

Technical Details

• Backend: FastAPI endpoint with Playwright integration

• Frontend: React + Vite with TanStack Query

• Event Loop: Uses ProactorEventLoop on Windows for subprocess support

• Stealth Mode: Anti-bot detection measures for better compatibility

• Error Handling: Graceful handling of sites that block automated access

See PLAYWRIGHT_SANDBOX_SETUP.md for detailed setup instructions.

🏗️ Project Structure

🚀 Deployment Options

Google Cloud Run (Recommended)

• Scalable - Auto-scales based on traffic

• Serverless - Pay only for what you use

• Managed - No infrastructure to manage

• Fast - Low latency with global CDN

See DEPLOY_CLOUD_RUN_WINDOWS.md for detailed instructions.

Docker (Local/Other Platforms)

📊 Performance

• P95 Latency: < 5 seconds for standard 3-step ReAct chains

• Scalability: Horizontal scaling on Cloud Run

• Uptime: 99.9% target (Cloud Run SLA)

• Throughput: Handles concurrent requests efficiently

🔒 Security

• API key authentication (optional)

• Environment variable management

• Secret Manager integration (Cloud Run)

• HTTPS by default (Cloud Run)

• CORS configuration

🤝 Contributing

We welcome contributions! Please see our contributing guidelines.

1. Fork the repository

2. Create a feature branch

3. Make your changes

4. Submit a pull request

📜 License

This project is licensed under the MIT License.

🔗 Links

• GitHub Repository: https://github.com/mcpmessenger/LangchainMCP

• Live Service: https://langchain-agent-mcp-server-554655392699.us-central1.run.app

• API Documentation: https://langchain-agent-mcp-server-554655392699.us-central1.run.app/docs

• Model Context Protocol: https://modelcontextprotocol.io/

🙏 Acknowledgments

• Built with LangChain

• Deployed on Google Cloud Run

• Uses FastAPI for the web framework

Status: ✅ Production-ready and deployed on Google Cloud Run