Get Started with MCP

Ever wished your Intelli flows could call your own code?
MCP (Model Context Protocol) makes that a one-liner.

Installation

pip install intelli[mcp]

That single extra [mcp] flag pulls in the small MCP SDK and its helpers.

Components at a Glance

Piece	What it does
MCP Server	Hosts the tools/functions you want to expose.
MCP Agent	A special Intelli agent that talks to that server.
Preprocessor (optional)	Picks data out of a previous task and updates the MCP agent's parameters before the call.

The Simplest Possible Flow

First, create a simple MCP server file named mcp_math_server.py:

# mcp_math_server.py
from mcp.server.fastmcp import FastMCP

# Create an MCP server
mcp = FastMCP("MathTools")

# Add a simple addition tool
@mcp.tool()
def add(a: int, b: int) -> int:
    """Add two numbers"""
    return a + b

if __name__ == "__main__":
    mcp.run(transport="stdio")

Now, create a client file that calls it:

# simple_mcp_client.py
import asyncio, sys
from intelli.flow.agents.agent import Agent
from intelli.flow.tasks.task import Task
from intelli.flow.input.task_input import TextTaskInput
from intelli.flow.flow import Flow
from intelli.flow.types import AgentTypes

# 1. MCP agent (local subprocess, stdio transport)
mcp_agent = Agent(
    agent_type=AgentTypes.MCP.value,
    provider="mcp",
    mission="Do simple math",
    model_params={
        "command": sys.executable,
        "args": ["mcp_math_server.py"],  # Points to the server we just created
        "tool": "add",
        "arg_a": 7,
        "arg_b": 8,
    }
)

# 2. Single task that just calls the tool
calc_task = Task(
    TextTaskInput("Calculate"),
    mcp_agent,
    log=True
)

# 3. One-task flow
flow = Flow({"calc": calc_task}, {"calc": []}, log=True)

# Run it
result = asyncio.run(flow.start())
print("Answer:", result["calc"]["output"])

Run this with:

python simple_mcp_client.py

That's all it takes! Your flow now calls your custom math function.

Check the richer example at: 🔗 Remote Server/Client 🔗 Dataframe MCP