GPT-OSS Browser MCP Server

A Rust-based MCP (Model Context Protocol) server that provides web browsing tools, optimized for ARM64 deployment on AWS Lambda.

🌟 Features

• 🌐 Web Browsing Tools: Search, open, and find functionality for web content
• 🏗️ ARM64 Optimized: Native ARM64 compilation for AWS Lambda Graviton processors
• 📋 MCP 2024-11-05 Compliant: Full protocol compliance with modern MCP standard
• 🔄 Session Management: Stateful browsing with in-memory session storage
• ⚡ Performance: Optimized for serverless deployment with fast cold starts
• 📊 Monitoring: CloudWatch integration with comprehensive logging and alarms
• 🔒 Security: CORS-enabled with session-based authentication

🛠️ Available Tools

1. Search (search)

Search for information on the web using DuckDuckGo and return formatted results with citations.

Parameters:

• query (required): Search query string
• topn (optional): Number of results to return (default: 10, max: 50)

Example:

{
  "query": "rust programming language",
  "topn": 5
}

2. Open (open)

Open a web page by URL and return its content converted to text with line numbers for citation.

Parameters:

• url (required): URL to open
• loc (optional): Starting line number (default: 0)
• num_lines (optional): Number of lines to show (-1 for all, default: -1)

Example:

{
  "url": "https://www.rust-lang.org",
  "loc": 0,
  "num_lines": 50
}

3. Find (find)

Find specific text patterns in the currently opened page or a specific URL.

Parameters:

• pattern (required): Text pattern to search for (case-insensitive)
• url (optional): URL of the page to search in (uses current page if omitted)

Example:

{
  "pattern": "memory safety",
  "url": "https://www.rust-lang.org"
}

🚀 Quick Start

Prerequisites

• Docker: For building the Lambda image
• AWS CLI: Configured with appropriate permissions
• AWS SAM CLI: For deployment
• Rust (optional): For local development

1. Clone and Build

# Navigate to the project directory
cd gpt-oss-mcp-server-rust

# Build the Docker image
./build-docker.sh

2. Deploy to AWS

# Deploy to production (default)
./deploy.sh

# Deploy to specific environment and region
./deploy.sh -e dev -r us-west-2

# Deploy with custom stack name
./deploy.sh -s my-mcp-server -e staging

# Get help with all options
./deploy.sh --help

Deploy Script Features:

• ✅ Automated prerequisites check (AWS CLI, SAM CLI, Docker)
• ✅ Native ARM64 compilation with Docker
• ✅ Smart stack management (handles failed deployments)
• ✅ Comprehensive endpoint testing after deployment
• ✅ Detailed deployment summary with performance metrics

Test Deployment:

# Run comprehensive deployment validation
./test-deployment.sh

3. Configure MCP Client

After deployment, you'll receive an MCP endpoint URL. Add it to your MCP client configuration:

{
	"mcpServers": {
		"gpt-oss-browser": {
			"command": "npx",
			"args": [
				"mcp-remote",
        "https://your-api-gateway-url/prod/mcp"
			]
		}
	}
}

🏗️ Architecture

Components

• Lambda Function: ARM64-optimized Rust application using lambda_runtime and axum
• API Gateway: HTTP endpoint with CORS support and proxy integration
• CloudWatch: Logging and monitoring with error/duration alarms
• Docker: Multi-stage build for ARM64 using Amazon Linux 2023

Session Management

The server maintains browsing sessions using in-memory storage:

• Each client gets a unique session ID via Mcp-Session-Id header
• Pages are cached within sessions for efficient access
• Session cleanup on termination
• Thread-safe concurrent access with RwLock

ARM64 Optimization

• Native ARM64 compilation: Compiles directly on ARM64 architecture using AL2023 base image
• Graviton2/3 processor optimization: Uses neoverse-n1 target CPU for

... View full README on GitHub