SIGN UPLOG IN
X (Twitter)LinkedInGitHubDiscordYouTube

How to integrate Replicate MCP with Autogen

Connect Autogen to Replicate MCP. Run stable diffusion to generate an image, list all your uploaded files on replicate, and more using natural language, with authentication handled for you.

Get started for freeGet a demo
Replicate logoReplicate
Api Key

Replicate is a cloud API platform for running AI models without managing infrastructure. Get scalable, production-ready access to the latest machine learning models in minutes.

33 Tools

Introduction

This guide walks you through connecting Replicate to AutoGen using the Composio tool router. By the end, you'll have a working Replicate agent that can run AI models with predictions, search public models, manage deployments, start training jobs, and upload files through natural language commands.

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

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

Also integrate Replicate with

Claude CodeCursorChatGPTClaude CoworkAntigravityOpenAI Agents SDKClaude Agent SDKCodexGrok BuildVS CodeOpenCodeOpenClawHermesAtomic AgentGoogle ADKLangChainVercel AI SDKMastra AILlamaIndexCrewAIPydantic AI

TL;DR

Here's what you'll learn:
  • Get and set up your OpenAI and Composio API keys
  • Install the required dependencies for Autogen and Composio
  • Initialize Composio and create a Tool Router session for Replicate
  • Wire that MCP URL into Autogen using McpWorkbench and StreamableHttpServerParams
  • Configure an Autogen AssistantAgent that can call Replicate tools
  • Run a live chat loop where you ask the agent to perform Replicate operations

What is AutoGen?

Autogen is a framework for building multi-agent conversational AI systems from Microsoft. It enables you to create agents that can collaborate, use tools, and maintain complex workflows.

Key features include:

  • Multi-Agent Systems: Build collaborative agent workflows
  • MCP Workbench: Native support for Model Context Protocol tools
  • Streaming HTTP: Connect to external services through streamable HTTP
  • AssistantAgent: Pre-built agent class for tool-using assistants

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

The Replicate MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Replicate account. It provides structured and secure access to your Replicate resources, so your agent can perform actions like running AI model predictions, managing files, browsing model collections, and retrieving model documentation on your behalf.

  • Run and manage AI model predictions: Easily instruct your agent to create, monitor, and manage predictions on any deployed Replicate model using custom input parameters.
  • Browse and discover model collections: Ask your agent to fetch and list available model collections or retrieve example predictions to explore what’s possible on Replicate.
  • Upload and organize files: Let your agent upload new files, list all stored files, or inspect file details to streamline your model workflows.
  • Access model metadata and documentation: Retrieve full model details, schemas, and markdown README docs for any model to help you choose and utilize the right model for your tasks.

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

What is Composio SDK?

Composio's Composio SDK 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 Composio SDK

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

How the Composio SDK works

The Composio SDK 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

Step by step08 STEPS
1

Prerequisites

You will need:

  • A Composio API key
  • An OpenAI API key (used by Autogen's OpenAIChatCompletionClient)
  • A Replicate account you can connect to Composio
  • Some basic familiarity with Autogen and Python async
2

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.
3

Install dependencies

bash
pip install composio python-dotenv
pip install autogen-agentchat autogen-ext-openai autogen-ext-tools

Install Composio, Autogen extensions, and dotenv.

What's happening:

  • composio connects your agent to Replicate via MCP
  • autogen-agentchat provides the AssistantAgent class
  • autogen-ext-openai provides the OpenAI model client
  • autogen-ext-tools provides MCP workbench support

4

Set up environment variables

bash
COMPOSIO_API_KEY=your-composio-api-key
OPENAI_API_KEY=your-openai-api-key
USER_ID=your-user-identifier@example.com

Create a .env file in your project folder.

What's happening:

  • COMPOSIO_API_KEY is required to talk to Composio
  • OPENAI_API_KEY is used by Autogen's OpenAI client
  • USER_ID is how Composio identifies which user's Replicate connections to use
5

Import dependencies and create Tool Router session

python
import asyncio
import os
from dotenv import load_dotenv
from composio import Composio

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import McpWorkbench, StreamableHttpServerParams

load_dotenv()

async def main():
    # Initialize Composio and create a Replicate session
    composio = Composio(api_key=os.getenv("COMPOSIO_API_KEY"))
    session = composio.create(
        user_id=os.getenv("USER_ID"),
        toolkits=["replicate"]
    )
    url = session.mcp.url
What's happening:
  • load_dotenv() reads your .env file
  • Composio(api_key=...) initializes the SDK
  • create(...) creates a Tool Router session that exposes Replicate tools
  • session.mcp.url is the MCP endpoint that Autogen will connect to
6

Configure MCP parameters for Autogen

python
# Configure MCP server parameters for Streamable HTTP
server_params = StreamableHttpServerParams(
    url=url,
    timeout=30.0,
    sse_read_timeout=300.0,
    terminate_on_close=True,
    headers={"x-api-key": os.getenv("COMPOSIO_API_KEY")}
)

Autogen expects parameters describing how to talk to the MCP server. That is what StreamableHttpServerParams is for.

What's happening:

  • url points to the Tool Router MCP endpoint from Composio
  • timeout is the HTTP timeout for requests
  • sse_read_timeout controls how long to wait when streaming responses
  • terminate_on_close=True cleans up the MCP server process when the workbench is closed
7

Create the model client and agent

python
# Create model client
model_client = OpenAIChatCompletionClient(
    model="gpt-5",
    api_key=os.getenv("OPENAI_API_KEY")
)

# Use McpWorkbench as context manager
async with McpWorkbench(server_params) as workbench:
    # Create Replicate assistant agent with MCP tools
    agent = AssistantAgent(
        name="replicate_assistant",
        description="An AI assistant that helps with Replicate operations.",
        model_client=model_client,
        workbench=workbench,
        model_client_stream=True,
        max_tool_iterations=10
    )

What's happening:

  • OpenAIChatCompletionClient wraps the OpenAI model for Autogen
  • McpWorkbench connects the agent to the MCP tools
  • AssistantAgent is configured with the Replicate tools from the workbench
8

Run the interactive chat loop

python
print("Chat started! Type 'exit' or 'quit' to end the conversation.\n")
print("Ask any Replicate related question or task to the agent.\n")

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

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

    if not user_input:
        continue

    print("\nAgent is thinking...\n")

    # Run the agent with streaming
    try:
        response_text = ""
        async for message in agent.run_stream(task=user_input):
            if hasattr(message, "content") and message.content:
                response_text = message.content

        # Print the final response
        if response_text:
            print(f"Agent: {response_text}\n")
        else:
            print("Agent: I encountered an issue processing your request.\n")

    except Exception as e:
        print(f"Agent: Sorry, I encountered an error: {str(e)}\n")
What's happening:
  • The script prompts you in a loop with You:
  • Autogen passes your input to the model, which decides which Replicate tools to call via MCP
  • agent.run_stream(...) yields streaming messages as the agent thinks and calls tools
  • Typing exit, quit, or bye ends the loop

Complete Code

Here's the complete code to get you started with Replicate and AutoGen:

python
import asyncio
import os
from dotenv import load_dotenv
from composio import Composio

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import McpWorkbench, StreamableHttpServerParams

load_dotenv()

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

    # Configure MCP server parameters for Streamable HTTP
    server_params = StreamableHttpServerParams(
        url=url,
        timeout=30.0,
        sse_read_timeout=300.0,
        terminate_on_close=True,
        headers={"x-api-key": os.getenv("COMPOSIO_API_KEY")}
    )

    # Create model client
    model_client = OpenAIChatCompletionClient(
        model="gpt-5",
        api_key=os.getenv("OPENAI_API_KEY")
    )

    # Use McpWorkbench as context manager
    async with McpWorkbench(server_params) as workbench:
        # Create Replicate assistant agent with MCP tools
        agent = AssistantAgent(
            name="replicate_assistant",
            description="An AI assistant that helps with Replicate operations.",
            model_client=model_client,
            workbench=workbench,
            model_client_stream=True,
            max_tool_iterations=10
        )

        print("Chat started! Type 'exit' or 'quit' to end the conversation.\n")
        print("Ask any Replicate related question or task to the agent.\n")

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

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

            if not user_input:
                continue

            print("\nAgent is thinking...\n")

            # Run the agent with streaming
            try:
                response_text = ""
                async for message in agent.run_stream(task=user_input):
                    if hasattr(message, 'content') and message.content:
                        response_text = message.content

                # Print the final response
                if response_text:
                    print(f"Agent: {response_text}\n")
                else:
                    print("Agent: I encountered an issue processing your request.\n")

            except Exception as e:
                print(f"Agent: Sorry, I encountered an error: {str(e)}\n")

if __name__ == "__main__":
    asyncio.run(main())

Conclusion

You now have an Autogen assistant wired into Replicate through Composio's Tool Router and MCP. From here you can:
  • Add more toolkits to the toolkits list, for example notion or hubspot
  • Refine the agent description to point it at specific workflows
  • Wrap this script behind a UI, Slack bot, or internal tool
Once the pattern is clear for Replicate, you can reuse the same structure for other MCP-enabled apps with minimal code changes.
TOOLS

Supported Tools

Every Replicate action and event your agent gets out of the box.

Get Account Information

Tool to get authenticated account information.

Cancel Prediction

Tool to cancel a prediction that is still running.

Get model collection

Tool to get a specific collection of models by its slug.

List model collections

Tool to list all collections of models.

Create Model

Tool to create a new Replicate model with specified owner, name, visibility, and hardware.

Create Prediction

Create a prediction for a Replicate Deployment using model inputs.

Create Deployment

Tool to create a new deployment with specified model, version, hardware, and scaling parameters.

Delete Deployment

Tool to delete a deployment from your account.

Get Deployment Details

Tool to get deployment details by owner and name.

List deployments

List deployments associated with the account, including their latest release configurations.

Create File

Tool to create or upload a file to Replicate.

Delete File

Delete a file by its ID.

Download File

Download a stored file using Replicate's signed download fields.

Get File Details

Tool to get details of a file by its ID.

List Files

Tool to retrieve a paginated list of uploaded files.

Get Prediction

Tool to get the status and output of a prediction by its ID.

List Available Hardware

Tool to list available hardware SKUs for models and deployments.

List model examples

List example predictions for a specific model, with cursor-based pagination.

Get Model Details

Get details of a specific model by owner and name.

List Public Models

Tool to list public models with pagination and sorting.

Create Model Prediction

Tool to create a prediction using an official Replicate model.

Get Model README

Tool to get the README content for a model in Markdown format.

Get Model Version

Tool to get a specific version of a model.

List Model Versions

List versions of a specific model, newest first, with cursor-based pagination.

Create Prediction

Tool to create a prediction to run a model by version ID.

List All Predictions

Tool to list all predictions for the authenticated user or organization with pagination.

Search Models and Collections

Tool to search for models, collections, and docs using text queries (beta).

Cancel Training

Tool to cancel an ongoing training operation in Replicate.

Create Training Job

Tool to create a training job for a specific model version.

Get Training

Get a training's current state, output, logs, and runtime metrics by ID.

List Training Jobs

Tool to list all training jobs for the authenticated user or organization with pagination.

Update Model Metadata

Update a model's metadata, including its description, README, and associated URLs.

Get Webhook Signing Secret

Tool to get the signing secret for the default webhook.

FRAMEWORKS

How to build Replicate MCP Agent with another framework

Claude Code logo

Claude Code

Use Replicate MCP with Claude Code

Cursor logo

Cursor

Use Replicate MCP with Cursor

ChatGPT logo

ChatGPT

Use Replicate MCP with ChatGPT

Claude Cowork logo

Claude Cowork

Use Replicate MCP with Claude Cowork

Antigravity logo

Antigravity

Use Replicate MCP with Antigravity

OpenAI Agents SDK logo

OpenAI Agents SDK

Use Replicate MCP with OpenAI Agents SDK

Claude Agent SDK logo

Claude Agent SDK

Use Replicate MCP with Claude Agent SDK

Codex logo

Codex

Use Replicate MCP with Codex

Grok Build logo

Grok Build

Use Replicate MCP with Grok Build

VS Code logo

VS Code

Use Replicate MCP with VS Code

OpenCode logo

OpenCode

Use Replicate MCP with OpenCode

OpenClaw logo

OpenClaw

Use Replicate MCP with OpenClaw

Hermes logo

Hermes

Use Replicate MCP with Hermes

Atomic Agent logo

Atomic Agent

Use Replicate MCP with Atomic Agent

Google ADK logo

Google ADK

Use Replicate MCP with Google ADK

LangChain logo

LangChain

Use Replicate MCP with LangChain

Vercel AI SDK logo

Vercel AI SDK

Use Replicate MCP with Vercel AI SDK

Mastra AI logo

Mastra AI

Use Replicate MCP with Mastra AI

LlamaIndex logo

LlamaIndex

Use Replicate MCP with LlamaIndex

CrewAI logo

CrewAI

Use Replicate MCP with CrewAI

Pydantic AI logo

Pydantic AI

Use Replicate MCP with Pydantic AI

FAQ

Frequently asked questions

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

Yes, you can. Autogen 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 Replicate tools.

Yes, absolutely. You can configure which Replicate 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.

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 Replicate data and credentials are handled as safely as possible.

Start with Replicate.It takes 30 seconds.

Managed auth, hosted MCP servers, and every Replicate tool your agent needs.Free to start.

Start building