mcp-agent-proxy
An MCP server that acts as a proxy to connect MCP clients to agent frameworks like Mastra and LangGraph, enabling agent discovery, dynamic server connections, and recursive agent networks.
README
MCP Agent Proxy
Connect any MCP client to any agent server - creating an "Internet of Agents" through simple, composable primitives.
Quick Start
Add this to your MCP client's configuration:
{
"mcpServers": {
"mcpAgentProxy": {
"command": "npx",
"args": ["mcp-agent-proxy@latest"]
}
}
}
That's it! The proxy lets your MCP client connect to different agent servers and exposes agent interactions as MCP tools. It automatically detects agent server types and adapts accordingly.
https://github.com/user-attachments/assets/9eaf1d36-298f-430f-b9e8-37e921cce2d1
What This Does
Instead of building complex protocols, we provide 5 simple tools that let agents discover, connect, and orchestrate across unlimited networks:
| Tool | Purpose |
|---|---|
listAgents |
Discover available agents across all servers |
describeAgent |
Get detailed agent capabilities for intelligent routing |
callAgent |
Execute any agent with smart conflict resolution |
connectServer |
Dynamically add new agent servers at runtime |
disconnectServer |
Remove dynamically connected servers |
Supported Server Types
- Mastra: Support for local and remote Mastra servers
- LangGraph: Support for local and remote LangGraph instances
- Extensible: Plugin architecture makes it easy to add support for other agent frameworks
The Network Effect
Agent servers can be MCP clients themselves, creating recursive agent networks:
Your MCP Client → MCP Proxy → Agent Server (Mastra/LangGraph) → Agents -> MCP Proxy -> Other Agent Servers → 🚀 Agents
One configuration line unlocks entire ecosystems of AI capabilities.
MCP Compliance & Features
✅ Fully MCP 2025-03-26 Compliant
- Complete implementation of the latest Model Context Protocol specification
🚀 Multi-Transport Support
- Stdio Transport: Standard MCP client integration (
npx mcp-agent-proxy) - Streamable HTTP: Modern HTTP-based transport with session management
- Legacy SSE: Backward compatibility with existing SSE implementations
- Automatic Detection: Seamlessly switches between transports based on client capabilities
Common Configurations
Multiple Servers (Mixed Types)
{
"mcpServers": {
"mcpAgentProxy": {
"command": "npx",
"args": ["mcp-agent-proxy"],
"env": {
"AGENT_SERVERS": "http://localhost:4111 http://localhost:2024"
}
}
}
}
Cloud + Local (Mixed Mastra and LangGraph)
{
"mcpServers": {
"mcpAgentProxy": {
"command": "npx",
"args": ["mcp-agent-proxy"],
"env": {
"AGENT_SERVERS": "https://my-mastra.vercel.app http://localhost:2024"
}
}
}
}
From Source (Development)
{
"mcpServers": {
"mcpAgentProxy": {
"command": "node",
"args": ["/path/to/mcp-agent-proxy/dist/mcp-server.js"],
"env": {
"AGENT_SERVERS": "http://localhost:4111"
}
}
}
}
Usage Examples
Basic Agent Call:
Can you call the weatherAgent to get the current weather in New York City?
Network Exploration
Explore the agents you're connected to
Smart Conflict Resolution:
I need to use the weatherAgent from server1 specifically, not the default one
Dynamic Network Expansion:
Connect to the ML specialists at https://ml-specialists.vercel.app and then use their modelTrainer agent
Environment Variables
| Variable | Default | Description |
|---|---|---|
AGENT_SERVERS |
(none) | Space/comma-separated server URLs (supports Mastra and LangGraph) |
MCP_SERVER_PORT |
3001 |
Proxy server port |
MCP_TRANSPORT |
http |
Transport method (stdio/http) |
Examples & Advanced Usage
Ready-to-use configurations in the examples/ directory:
mcp.json- Zero setupmulti-server-config.json- Multiple local serversvercel-config.json- Cloud deployment
For advanced configuration options, see CONFIGURATION.md.
Troubleshooting
NPX Issues:
npx clear-npx-cache
npx mcp-agent-proxy@latest
Port Conflicts:
"env": { "MCP_SERVER_PORT": "3002" }
Connection Issues:
- Ensure agent servers are running and accessible
- Check firewall settings and server URLs
- For LangGraph: Default port is usually 2024 (
langgraph dev) - For Mastra: Default port is usually 4111
Debug Mode:
DEBUG=mastra:* npx mcp-agent-proxy
Development & Testing
Validation Workflows
We've implemented comprehensive validation workflows to ensure code quality and MCP protocol compliance:
Quick Commands
# Quick validation (build + test + lint + format)
pnpm validate:quick
# Full validation (includes integration tests + health checks)
pnpm validate:full
# CI validation (includes coverage + dead code analysis)
pnpm validate:ci
Individual Commands
# Build and test
pnpm build
pnpm test
pnpm test:integration
# Code quality
pnpm lint
pnpm format
pnpm format:check
MCP Protocol Testing
# Test MCP protocol compliance
pnpm mcp:test
# Test security features
pnpm security:test
# Manual server testing
pnpm start # Start server
pnpm check # Check running server (requires jq)
Test Coverage
- 355 total tests covering all components
- Unit tests for all tools, plugins, and configuration
- Integration tests for MCP protocol compliance
- Security tests for origin validation and session management
- Health checks for server monitoring
Available Scripts
Use pnpm run to see all available scripts, or use these common ones:
pnpm dev # Build and start development server
pnpm test # Run all tests
pnpm test:watch # Run tests in watch mode
pnpm build # Build for production
pnpm lint # Run ESLint
pnpm format # Format code with Prettier
Contributing
We welcome contributions! Please see our Contributing Guide for details.
License
MIT License - see LICENSE file for details.
<div align="center">
📖 Configuration Guide • 🚀 Examples • 🐛 Issues • 💬 Discussions
Building the Internet of Agents, one connection at a time.
</div>
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。