Skip to content

MCP Integration

Connect your agents to Model Context Protocol (MCP) servers to access external tools dynamically.

What is MCP?

Model Context Protocol (MCP) is Anthropic's open standard for connecting AI assistants to data sources and tools.

Benefits

  • Connect to external tool servers
  • Auto-discover available tools
  • Secure tool execution
  • Works with any MCP-compatible server

Installation

pip install react-agent-framework[mcp]

Basic Usage

from react_agent_framework import ReactAgent

# Create agent
agent = ReactAgent(name="MCP Agent")

# Connect to filesystem MCP server
server_id = agent.add_mcp_server(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-filesystem", "/tmp"],
    name="filesystem"
)

# All tools from the server are now available!
answer = agent.run("List files in the /tmp directory")

Filesystem Server

Access local files and directories:

agent.add_mcp_server(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"],
    name="filesystem"
)

GitHub Server

Interact with GitHub repositories:

agent.add_mcp_server(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-github"],
    env={"GITHUB_TOKEN": "ghp_your_token"},
    name="github"
)

PostgreSQL Server

Query databases:

agent.add_mcp_server(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-postgres"],
    env={"POSTGRES_CONNECTION_STRING": "postgresql://..."},
    name="postgres"
)

Puppeteer Server

Browser automation:

agent.add_mcp_server(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-puppeteer"],
    name="puppeteer"
)

Managing Servers

List Connected Servers

servers = agent.list_mcp_servers()
for server in servers:
    print(f"{server['name']}: {server['num_tools']} tools")

List Available Tools

tools = agent.list_mcp_tools()
for tool_desc in tools:
    print(tool_desc)

Disconnect from Server

agent.disconnect_mcp_server(server_id)

Using Config Files

Create mcp_config.json:

{
  "servers": [
    {
      "name": "filesystem",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"],
      "auto_connect": true
    },
    {
      "name": "github",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "ghp_..."
      },
      "auto_connect": false
    }
  ]
}

Load config:

from react_agent_framework.mcp.config import MCPConfigManager

config_manager = MCPConfigManager("mcp_config.json")

for server_name in config_manager.list_servers():
    server_config = config_manager.get_server(server_name)
    agent.add_mcp_server(
        command=server_config.command,
        args=server_config.args,
        env=server_config.env,
        name=server_config.name
    )

Complete Example

from react_agent_framework import ReactAgent

# Create agent
agent = ReactAgent(
    name="Multi-Source Agent",
    description="Agent with access to multiple data sources",
    provider="gpt-4o-mini"
)

# Connect to filesystem
fs_server = agent.add_mcp_server(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-filesystem", "."],
    name="filesystem"
)

# Connect to GitHub (if token available)
try:
    gh_server = agent.add_mcp_server(
        command="npx",
        args=["-y", "@modelcontextprotocol/server-github"],
        env={"GITHUB_TOKEN": "ghp_..."},
        name="github"
    )
except:
    print("GitHub server not configured")

# List all available tools
print("\nAvailable MCP Tools:")
for tool in agent.list_mcp_tools():
    print(f"  - {tool}")

# Use the agent
tasks = [
    "List all Python files in current directory",
    "Read the contents of README.md",
]

for task in tasks:
    print(f"\nTask: {task}")
    result = agent.run(task)
    print(f"Result: {result}")

# Cleanup
agent.disconnect_mcp_server(fs_server)

Creating Custom MCP Servers

You can create your own MCP servers. See the MCP Python SDK documentation for details.

Example custom server:

from mcp import Server, Tool

server = Server("my-custom-server")

@server.tool()
async def my_custom_tool(param: str) -> str:
    """My custom tool"""
    return f"Processed: {param}"

if __name__ == "__main__":
    server.run()

Use it:

agent.add_mcp_server(
    command="python",
    args=["my_custom_server.py"],
    name="custom"
)

API Reference

For detailed API documentation, see:

Learn More