Skip to content

Using MCP Tools in RailTracks

Overview

Quick Summary

RailTracks makes it easy to use any MCP-compatible tool with your agents. Just connect to an MCP server, get the tools, and start using them!

RailTracks supports seamless integration with Model Context Protocol (MCP), allowing you to use any MCP-compatible tool as a native RailTracks Tool. This means you can connect your agents to a wide variety of external tools and data sources—without having to implement the tool logic yourself.

RailTracks handles the discovery and invocation of MCP tools, so you can focus on building intelligent agents.

Prerequisites

Before You Begin

Make sure you have the following set up before using MCP tools:

  • RailTracks Framework installed (pip install railtracks[core])
  • MCP package set up - Every MCP tool has different requirements (see specific tool documentation)
  • Authentication credentials - Many MCP tools require API keys or OAuth tokens

Connecting to MCP Server Types

RailTracks supports two types of MCP servers

Remote HTTP Servers

Use MCPHttpParams for connecting to remote MCP servers:

import railtracks as rt


# Connect to a remote MCP server
fetch_server = rt.connect_mcp(
    rt.MCPHttpParams(
        url="https://remote.mcpservers.org/fetch/mcp",
        # Optional: Add authentication headers if needed
        headers={"Authorization": f"Bearer {'<API_KEY>'}"},
    )
)

Local Stdio Servers

Use MCPStdioParams for running local MCP servers:

import railtracks as rt

# Run a local MCP server (Time server example)
time_server = rt.connect_mcp(
    rt.MCPStdioParams(
        command="npx",
        args=["mcp-server-time"]  # or other command to run the server
    )
)

Using MCP Tools with RailTracks Agents

Once you've connected to an MCP server, you can use the tools with your RailTracks agents:

import railtracks as rt

# Run a local MCP server (Time server example)
time_server = rt.connect_mcp(
    rt.MCPStdioParams(
        command="npx",
        args=["mcp-server-time"]  # or other command to run the server
    )
)

Common MCP Server Examples

Fetch Server (URL Content Retrieval) 

fetch_server = rt.connect_mcp(
    rt.MCPHttpParams(url="https://remote.mcpservers.org/fetch/mcp")
)
Guide: Websearch Server

GitHub Server 
github_server = rt.connect_mcp(
    rt.MCPHttpParams(
        url="https://api.githubcopilot.com/mcp/",
        headers={
            "Authorization": f"Bearer {'<GITHUB_PAT_TOKEN>'}",
        },
    )
)

Guide: Github Server

Warning

If you fail to provde the correct PAT you will see the following error:

Exception in thread Thread-1 (_thread_main):

Traceback (most recent call last):

File "C:\Users\rc\.venv\lib\site-packages\anyio\streams\memory.py", line 111, in receive
Notion Server 

import json

notion_server = rt.connect_mcp(
    rt.MCPStdioParams(
        command="npx",
        args=["-y", "@notionhq/notion-mcp-server"],
        env={
            "OPENAPI_MCP_HEADERS": json.dumps({
                "Authorization": f"Bearer {'<NOTION_API_TOKEN>'}",
                "Notion-Version": "2022-06-28"
            })
        },
    )
)
Guide: Notion Server

Combining Multiple MCP Tools

You can combine tools from different MCP's into one single agent. 

# You can combine the tools from multiple MCP servers
all_tools = notion_server.tools + github_server.tools + fetch_server.tools

# Create an agent that can use all tools
super_agent = rt.agent_node(
    tool_nodes=all_tools,
    name="Multi-Tool Agent",
    system_message="Use the appropriate tools to complete tasks.",
    llm=rt.llm.OpenAILLM("gpt-4o"),
)

Tool-Specific Guides

For detailed setup and usage instructions for specific MCP tools: