How to integrate Kaggle MCP with CrewAI

BOOK A CALLGET STARTED FOR FREE
Framework Integration Gradient
Kaggle Logo
CrewAI Logo
divider

Introduction

This guide walks you through connecting Kaggle to CrewAI using the Composio tool router. By the end, you'll have a working Kaggle agent that can download data files for the titanic competition, create a new version of my covid-19 dataset, check processing status of my uploaded dataset, submit my predictions to the house prices competition through natural language commands.

This guide will help you understand how to give your CrewAI agent real control over a Kaggle account through Composio's Kaggle MCP server.

Before we dive in, let's take a quick look at the key ideas and tools involved.

TL;DR

Here's what you'll learn:
  • Get a Composio API key and configure your Kaggle connection
  • Set up CrewAI with an MCP enabled agent
  • Create a Tool Router session or standalone MCP server for Kaggle
  • Build a conversational loop where your agent can execute Kaggle operations

What is CrewAI?

CrewAI is a powerful framework for building multi-agent AI systems. It provides primitives for defining agents with specific roles, creating tasks, and orchestrating workflows through crews.

Key features include:

  • Agent Roles: Define specialized agents with specific goals and backstories
  • Task Management: Create tasks with clear descriptions and expected outputs
  • Crew Orchestration: Combine agents and tasks into collaborative workflows
  • MCP Integration: Connect to external tools through Model Context Protocol

What is the Kaggle MCP server, and what's possible with it?

The Kaggle MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Kaggle account. It provides structured and secure access to your Kaggle datasets, competitions, and configurations, so your agent can perform actions like downloading competition data, creating datasets, submitting entries, and managing dataset versions on your behalf.

  • Competition data access and download: Let your agent fetch and download competition datasets quickly by specifying a competition ID, so you always have the latest files for analysis.
  • Automated dataset creation and management: Have your agent create new Kaggle datasets, update metadata, and publish new dataset versions seamlessly, streamlining the process of sharing your work with the community.
  • Competition entry submission: Empower your agent to submit competition entries automatically once your solution is ready and uploaded, helping you participate in challenges without manual hassle.
  • Configuration management and setup: Allow your agent to initialize, locate, and update Kaggle API configuration files and keys, ensuring smooth and authenticated operations every time.
  • Dataset status monitoring: Ask your agent to check the status of uploaded datasets or processing jobs, so you always know when your data is ready for use or public sharing.

Supported Tools & Triggers

Tools
Download competition data filesTool to download competition data files.
Initialize Kaggle ConfigurationTool to initialize Kaggle API client configuration.
Dataset CreateTool to create a new Kaggle dataset with full metadata.
Get Dataset StatusTool to get the status of a dataset upload or processing job.
Create Dataset VersionTool to create a new dataset version on Kaggle.
Submit Competition EntryTool to submit an entry to a Kaggle competition.
Get Kaggle Config DirectoryTool to retrieve the directory of the Kaggle API configuration file.
Get Kaggle Config FileTool to retrieve the filename of the Kaggle API configuration file.
List Kaggle Configuration KeysTool to list local Kaggle API configuration keys.
Get Kaggle Config PathTool to retrieve local Kaggle API configuration file path.
Reset Kaggle ConfigurationTool to reset local Kaggle CLI configuration to defaults.
Set Kaggle ConfigurationTool to set a Kaggle CLI configuration parameter.
Unset Kaggle ConfigurationTool to unset a Kaggle CLI configuration parameter.
View Kaggle ConfigurationTool to view local Kaggle API configuration.
Kaggle Dataset InitTool to initialize a dataset-metadata.
List Kaggle Dataset FilesTool to list files in a Kaggle dataset.
Kaggle Kernel InitTool to initialize a kernel-metadata.
Download kernel outputTool to download the output of a Kaggle kernel.
Get Kernel StatusTool to get the status of a Kaggle kernel run.
List Kaggle DatasetsTool to list Kaggle datasets with filters and pagination.

What is the Composio tool router, and how does it fit here?

What is Tool Router?

Composio's Tool Router helps agents find the right tools for a task at runtime. You can plug in multiple toolkits (like Gmail, HubSpot, and GitHub), and the agent will identify the relevant app and action to complete multi-step workflows. This can reduce token usage and improve the reliability of tool calls. Read more here: Getting started with Tool Router

The tool router generates a secure MCP URL that your agents can access to perform actions.

How the Tool Router works

The Tool Router follows a three-phase workflow:

  1. Discovery: Searches for tools matching your task and returns relevant toolkits with their details.
  2. Authentication: Checks for active connections. If missing, creates an auth config and returns a connection URL via Auth Link.
  3. Execution: Executes the action using the authenticated connection.

Step-by-step Guide

Prerequisites

Before starting, make sure you have:
  • Python 3.9 or higher
  • A Composio account and API key
  • A Kaggle connection authorized in Composio
  • An OpenAI API key for the CrewAI LLM
  • Basic familiarity with Python

Getting API Keys for OpenAI and Composio

OpenAI API Key
  • Go to the OpenAI dashboard and create an API key. You'll need credits to use the models, or you can connect to another model provider.
  • Keep the API key safe.
Composio API Key
  • Log in to the Composio dashboard.
  • Navigate to your API settings and generate a new API key.
  • Store this key securely as you'll need it for authentication.

Install dependencies

bash
pip install composio crewai crewai-tools python-dotenv
What's happening:
  • composio connects your agent to Kaggle via MCP
  • crewai provides Agent, Task, Crew, and LLM primitives
  • crewai-tools includes MCP helpers
  • python-dotenv loads environment variables from .env

Set up environment variables

bash
COMPOSIO_API_KEY=your_composio_api_key_here
USER_ID=your_user_id_here
OPENAI_API_KEY=your_openai_api_key_here

Create a .env file in your project root.

What's happening:

  • COMPOSIO_API_KEY authenticates with Composio
  • USER_ID scopes the session to your account
  • OPENAI_API_KEY lets CrewAI use your chosen OpenAI model

Import dependencies

python
from crewai import Agent, Task, Crew, LLM
from crewai_tools import MCPServerAdapter  # optional import if you plan to adapt tools
from composio import Composio
from dotenv import load_dotenv
import os
from crewai.mcp import MCPServerHTTP

load_dotenv()
What's happening:
  • CrewAI classes define agents and tasks, and run the workflow
  • MCPServerHTTP connects the agent to an MCP endpoint
  • Composio will give you a short lived Kaggle MCP URL

Create a Composio Tool Router session for Kaggle

python
composio = Composio(api_key=os.getenv("COMPOSIO_API_KEY"))
session = composio.create(
    user_id=os.getenv("USER_ID"),
    toolkits=["kaggle"],
)
url = session.mcp.url
What's happening:
  • You create a Kaggle only session through Composio
  • Composio returns an MCP HTTP URL that exposes Kaggle tools

Configure the LLM

python
llm = LLM(
    model="gpt-5-mini",
    api_key=os.getenv("OPENAI_API_KEY"),
)
What's happening:
  • CrewAI will call this LLM for planning and responses
  • You can swap in a different model if needed

Attach the MCP server and create the agent

python
toolkit_agent = Agent(
    role="Kaggle Assistant",
    goal="Help users interact with Kaggle through natural language commands",
    backstory=(
        "You are an expert assistant with access to Kaggle tools. "
        "You can perform various Kaggle operations on behalf of the user."
    ),
    mcps=[
        MCPServerHTTP(
            url=url,
            streamable=True,
            cache_tools_list=True,
            headers={"x-api-key": os.getenv("COMPOSIO_API_KEY")},
        ),
    ],
    llm=llm,
    verbose=True,
    max_iter=10,
)
What's happening:
  • MCPServerHTTP connects the agent to the Kaggle MCP endpoint
  • cache_tools_list saves a tools catalog for faster subsequent runs
  • verbose helps you see what the agent is doing

Add a REPL loop with Task and Crew

python
print("Chat started! Type 'exit' or 'quit' to end.\n")
print("Try asking the agent to perform Kaggle operations.\n")

conversation_context = ""

while True:
    user_input = input("You: ").strip()

    if user_input.lower() in ["exit", "quit", "bye"]:
        print("\nGoodbye!")
        break

    if not user_input:
        continue

    conversation_context += f"\nUser: {user_input}\n"
    print("\nAgent is thinking...\n")

    task = Task(
        description=(
            f"Based on the conversation history:\n{conversation_context}\n\n"
            f"Current user request: {user_input}\n\n"
            f"Please help the user with their Kaggle related request."
        ),
        expected_output="A helpful response addressing the user's request",
        agent=toolkit_agent,
    )

    crew = Crew(
        agents=[toolkit_agent],
        tasks=[task],
        verbose=False,
    )

    result = crew.kickoff()
    response = str(result)

    conversation_context += f"Agent: {response}\n"
    print(f"Agent: {response}\n")
What's happening:
  • You build a simple chat loop and keep a running context
  • Each user turn becomes a Task handled by the same agent
  • Crew executes the task and returns a response

Run the application

python
if __name__ == "__main__":
    main()
What's happening:
  • Standard Python entry point so you can run python crewai_kaggle_agent.py

Complete Code

Here's the complete code to get you started with Kaggle and CrewAI:

python
# file: crewai_kaggle_agent.py
from crewai import Agent, Task, Crew, LLM
from crewai_tools import MCPServerAdapter  # optional
from composio import Composio
from dotenv import load_dotenv
import os
from crewai.mcp import MCPServerHTTP

load_dotenv()

def main():
    # Initialize Composio and create a Kaggle session
    composio = Composio(api_key=os.getenv("COMPOSIO_API_KEY"))
    session = composio.create(
        user_id=os.getenv("USER_ID"),
        toolkits=["kaggle"],
    )
    url = session.mcp.url

    # Configure LLM
    llm = LLM(
        model="gpt-5-mini",
        api_key=os.getenv("OPENAI_API_KEY"),
    )

    # Create Kaggle assistant agent
    toolkit_agent = Agent(
        role="Kaggle Assistant",
        goal="Help users interact with Kaggle through natural language commands",
        backstory=(
            "You are an expert assistant with access to Kaggle tools. "
            "You can perform various Kaggle operations on behalf of the user."
        ),
        mcps=[
            MCPServerHTTP(
                url=url,
                streamable=True,
                cache_tools_list=True,
                headers={"x-api-key": os.getenv("COMPOSIO_API_KEY")},
            ),
        ],
        llm=llm,
        verbose=True,
        max_iter=10,
    )

    print("Chat started! Type 'exit' or 'quit' to end the conversation.\n")
    print("Try asking the agent to perform Kaggle operations.\n")

    conversation_context = ""

    while True:
        user_input = input("You: ").strip()

        if user_input.lower() in ["exit", "quit", "bye"]:
            print("\nGoodbye!")
            break

        if not user_input:
            continue

        conversation_context += f"\nUser: {user_input}\n"
        print("\nAgent is thinking...\n")

        task = Task(
            description=(
                f"Based on the conversation history:\n{conversation_context}\n\n"
                f"Current user request: {user_input}\n\n"
                f"Please help the user with their Kaggle related request."
            ),
            expected_output="A helpful response addressing the user's request",
            agent=toolkit_agent,
        )

        crew = Crew(
            agents=[toolkit_agent],
            tasks=[task],
            verbose=False,
        )

        result = crew.kickoff()
        response = str(result)

        conversation_context += f"Agent: {response}\n"
        print(f"Agent: {response}\n")

if __name__ == "__main__":
    main()

Conclusion

You now have a CrewAI agent connected to Kaggle through Composio's Tool Router. The agent can perform Kaggle operations through natural language commands. Next steps:
  • Add role-specific instructions to customize agent behavior
  • Plug in more toolkits for multi-app workflows
  • Chain tasks for complex multi-step operations

How to build Kaggle MCP Agent with another framework

OpenAI Agents SDK

OpenAI Agents SDK

Use Kaggle MCP with OpenAI Agents SDK

Claude Agent SDK

Claude Agent SDK

Use Kaggle MCP with Claude Agent SDK

Google ADK

Google ADK

Use Kaggle MCP with Google ADK

LangChain

LangChain

Use Kaggle MCP with LangChain

Vercel AI SDK

Vercel AI SDK

Use Kaggle MCP with Vercel AI SDK

Mastra AI

Mastra AI

Use Kaggle MCP with Mastra AI

LlamaIndex

LlamaIndex

Use Kaggle MCP with LlamaIndex

FAQ

What are the differences in Tool Router MCP and Kaggle MCP?

With a standalone Kaggle MCP server, the agents and LLMs can only access a fixed set of Kaggle tools tied to that server. However, with the Composio Tool Router, agents can dynamically load tools from Kaggle and many other apps based on the task at hand, all through a single MCP endpoint.

Can I use Tool Router MCP with CrewAI?

Yes, you can. CrewAI fully supports MCP integration. You get structured tool calling, message history handling, and model orchestration while Tool Router takes care of discovering and serving the right Kaggle tools.

Can I manage the permissions and scopes for Kaggle while using Tool Router?

Yes, absolutely. You can configure which Kaggle scopes and actions are allowed when connecting your account to Composio. You can also bring your own OAuth credentials or API configuration so you keep full control over what the agent can do.

How safe is my data with Composio Tool Router?

All sensitive data such as tokens, keys, and configuration is fully encrypted at rest and in transit. Composio is SOC 2 Type 2 compliant and follows strict security practices so your Kaggle data and credentials are handled as safely as possible.

Used by agents from

Context
ASU
Letta
glean
HubSpot
Agent.ai
Altera
DataStax
Entelligence
Rolai
Context
ASU
Letta
glean
HubSpot
Agent.ai
Altera
DataStax
Entelligence
Rolai
Context
ASU
Letta
glean
HubSpot
Agent.ai
Altera
DataStax
Entelligence
Rolai

Never worry about agent reliability

We handle tool reliability, observability, and security so you never have to second-guess an agent action.

Book a call↗Get started for free