Using GitHub MCP Server with RequestCompletion

To use the GitHub MCP server with RT, use the from_mcp_server utility to load tools directly from the MCP server. A valid GitHub Personal Access Token (PAT) is required, which in this example is provided via an environment variable.

import os
from railtracks.rt_mcp import MCPHttpParams
from railtracks import connect_mcp

server = connect_mcp(
    MCPHttpParams(
        url="https://api.githubcopilot.com/mcp/",
        headers={
            "Authorization": f"Bearer {os.getenv('GITHUB_PAT_TOKEN')}",
        },
    )
)
tools = server.tools

At this point, the tools can be used the same as any other RT tool. See the following code as a simple example.

import railtracks as rt
import asyncio

agent = rt.agent_node(
    # tool_nodes={*tools},    # Uncomment this line to use the tools
    system_message="""You are a GitHub Copilot agent that can interact with GitHub repositories.""",
    llm=rt.llm.OpenAILLM("gpt-4o"),
)

user_prompt = """Tell me about the RailtownAI/rc repository on GitHub."""

async def call_node():
    with rt.Session():
        result = await rt.call(agent, user_prompt)

    print(result.content)

# asyncio.run(call_node())