Back to Llama Index

MCP ToolSpec

llama-index-integrations/tools/llama-index-tools-mcp/README.md

0.14.217.1 KB
Original Source

MCP ToolSpec

This tool connects to MCP Servers and allows an Agent to call the tools provided by MCP Servers.

This idea is migrated from Integrate MCP Tools into LlamaIndex.

Installation

bash
pip install llama-index-tools-mcp

Usage

Usage is as simple as connecting to an MCP Server and getting the tools.

python
from llama_index.tools.mcp import BasicMCPClient, McpToolSpec

# We consider there is a mcp server running on 127.0.0.1:8000, or you can use the mcp client to connect to your own mcp server.
mcp_client = BasicMCPClient("http://127.0.0.1:8000/sse")
mcp_tool_spec = McpToolSpec(
    client=mcp_client,
    # Optional: Filter the tools by name
    # allowed_tools=["tool1", "tool2"],
    # Optional: Include resources in the tool list
    # include_resources=True,
)

# sync
tools = mcp_tool_spec.to_tool_list()

# async
tools = await mcp_tool_spec.to_tool_list_async()

Then you can use the tools in your agent!

python
from llama_index.core.agent.workflow import FunctionAgent
from llama_index.llms.openai import OpenAI

agent = FunctionAgent(
    name="Agent",
    description="Some description",
    llm=OpenAI(model="gpt-4o"),
    tools=tools,
    system_prompt="You are a helpful assistant.",
)

resp = await agent.run("What is the weather in Tokyo?")

Helper Functions

This package also includes several helper functions for working with MCP Servers.

workflow_as_mcp

This function converts a Workflow to an MCP app.

python
from llama_index.core.workflow import (
    Context,
    Workflow,
    Event,
    StartEvent,
    StopEvent,
    step,
)
from llama_index.tools.mcp import workflow_as_mcp


class RunEvent(StartEvent):
    msg: str


class InfoEvent(Event):
    msg: str


class LoudWorkflow(Workflow):
    """Useful for converting strings to uppercase and making them louder."""

    @step
    def step_one(self, ctx: Context, ev: RunEvent) -> StopEvent:
        ctx.write_event_to_stream(InfoEvent(msg="Hello, world!"))

        return StopEvent(result=ev.msg.upper() + "!")


workflow = LoudWorkflow()

mcp = workflow_as_mcp(workflow, start_event_model=RunEvent)

Then, you can launch the MCP server (assuming you have the mcp[cli] extra installed):

bash
mcp dev script.py

get_tools_from_mcp_url / aget_tools_from_mcp_url

This function get a list of FunctionTools from an MCP server or command.

python
from llama_index.tools.mcp import (
    get_tools_from_mcp_url,
    aget_tools_from_mcp_url,
)

tools = get_tools_from_mcp_url("http://127.0.0.1:8000/sse")

# async
tools = await get_tools_from_mcp_url("http://127.0.0.1:8000/sse")

MCP Client Usage

The BasicMCPClient provides comprehensive access to MCP server capabilities beyond just tools.

Basic Client Operations

python
from llama_index.tools.mcp import BasicMCPClient

# Connect to an MCP server using different transports
http_client = BasicMCPClient("https://example.com/mcp")  # Streamable HTTP
sse_client = BasicMCPClient("https://example.com/sse")  # Server-Sent Events
local_client = BasicMCPClient("python", args=["server.py"])  # stdio

# List available tools
tools = await http_client.list_tools()

# Call a tool
result = await http_client.call_tool("calculate", {"x": 5, "y": 10})

# List available resources
resources = await http_client.list_resources()

# Read a resource
content, mime_type = await http_client.read_resource("config://app")

# List available prompts
prompts = await http_client.list_prompts()

# Get a prompt
prompt_result = await http_client.get_prompt("greet", {"name": "World"})

OAuth Authentication

The client supports OAuth 2.0 authentication for connecting to protected MCP servers:

python
from llama_index.tools.mcp import BasicMCPClient

# Simple authentication with in-memory token storage
client = BasicMCPClient.with_oauth(
    "https://api.example.com/mcp",
    client_name="My App",
    redirect_uris=["http://localhost:3000/callback"],
    # Function to handle the redirect URL (e.g., open a browser)
    redirect_handler=lambda url: print(f"Please visit: {url}"),
    # Function to get the authorization code from the user
    callback_handler=lambda: (input("Enter the code: "), None),
)

# Use the authenticated client
tools = await client.list_tools()

For production use, you can implement a custom token storage:

python
from llama_index.tools.mcp import BasicMCPClient
from mcp.client.auth import TokenStorage
from mcp.shared.auth import OAuthToken, OAuthClientInformationFull
import json
import os


class FileTokenStorage(TokenStorage):
    """Store OAuth tokens in a file."""

    def __init__(self, file_path: str):
        self.file_path = file_path
        self._client_info: Optional[OAuthClientInformationFull] = None

    async def get_tokens(self):
        if not os.path.exists(self.file_path):
            return None
        with open(self.file_path, "r") as f:
            data = json.load(f)
            return OAuthToken(**data.get("tokens", {}))

    async def set_tokens(self, tokens):
        data = {}
        if os.path.exists(self.file_path):
            with open(self.file_path, "r") as f:
                data = json.load(f)
        data["tokens"] = tokens.__dict__
        with open(self.file_path, "w") as f:
            json.dump(data, f)

    async def get_client_info(self) -> Optional[OAuthClientInformationFull]:
        """Get the stored client information."""
        return self._client_info

    async def set_client_info(
        self, client_info: OAuthClientInformationFull
    ) -> None:
        """Store client information."""
        self._client_info = client_info


# Use custom storage
client = BasicMCPClient.with_oauth(
    "https://api.example.com/mcp",
    client_name="My App",
    redirect_uris=["http://localhost:3000/callback"],
    redirect_handler=lambda url: print(f"Please visit: {url}"),
    callback_handler=lambda: (input("Enter the code: "), None),
    token_storage=FileTokenStorage("tokens.json"),
)

Use a custom HTTP Client

The MCP Client can use a custom httpx.AsyncClient instance. This feature is only available for Streamable HTTP transport.

python
httpx_async_client = AsyncClient(
    # to avoid self-signed certificates failures
    verify=false
)

client = BasicMCPClient(
    url="https://example.com/mcp", http_client=httpx_async_client
)
client_with_oauth = BasicMCPClient.with_oauth(
    "https://api.example.com/mcp",
    client_name="My App",
    redirect_uris=["http://localhost:3000/callback"],
    # Function to handle the redirect URL (e.g., open a browser)
    redirect_handler=lambda url: print(f"Please visit: {url}"),
    # Function to get the authorization code from the user
    callback_handler=lambda: (input("Enter the code: "), None),
    http_client=httpx_async_client,
)

Notebook Example

This tool has a more extensive example usage documented in a Jupyter notebook here.

This tool is designed to be used as a way to call the tools provided by MCP Servers.