mcp_agent

MCP Agent Example

This example demonstrates how to use MCP (Model Context Protocol) tools with LangGraphGo agents.

Prerequisites

1. Configure MCP Servers: Create a ~/.claude.json file with your MCP server configurations:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/username/Documents"]
    },
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"],
      "env": {
        "BRAVE_API_KEY": "your-brave-api-key"
      }
    }
  }
}

1. Install MCP Servers: The example uses npm packages, so make sure you have Node.js installed:

npm install -g @modelcontextprotocol/server-filesystem
npm install -g @modelcontextprotocol/server-brave-search

1. Set OpenAI API Key:

export OPENAI_API_KEY="your-openai-api-key"

Running the Example

cd examples/mcp_agent
go run main.go

What It Does

1. Loads MCP Configuration: Reads ~/.claude.json to discover available MCP servers
2. Connects to Servers: Establishes connections to all configured MCP servers
3. Retrieves Tools: Fetches available tools from each server
4. Creates Agent: Sets up a LangGraphGo agent with MCP tools
5. Processes Query: Runs a sample query that may use MCP tools
6. Displays Results: Shows the agent's response and tool usage

Available MCP Servers

Here are some popular MCP servers you can use:

Filesystem

{
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/directory"]
}

Brave Search

{
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-brave-search"],
  "env": {
    "BRAVE_API_KEY": "your-api-key"
  }
}

PostgreSQL

{
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
}

GitHub

{
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"],
  "env": {
    "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token"
  }
}

Slack

{
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-slack"],
  "env": {
    "SLACK_BOT_TOKEN": "your-token"
  }
}

Customization

You can modify the example to:

• Use different queries
• Add custom MCP servers
• Implement streaming responses
• Add memory/state management
• Use different LLM models

Troubleshooting

Connection Failures

If you see "Failed to connect to MCP server" errors:

• Check that the MCP server command is installed and accessible
• Verify environment variables are set correctly
• Ensure paths in the config are absolute and exist

Tool Not Found

If tools aren't appearing:

• Run npx <server-package> directly to test the server
• Check server logs in stderr
• Verify the server supports the MCP protocol version

API Key Issues

Make sure all required API keys are set:

• OPENAI_API_KEY for the LLM
• Server-specific keys (BRAVE_API_KEY, GITHUB_TOKEN, etc.)