How to integrate Firecrawl MCP with OpenAI Agents SDK

Trusted by
AWS
Glean
Zoom
Airtable
GET STARTED FOR FREEGET A DEMO

30 min · no commitment · see it on your stack

Firecrawl logo
OpenAI Agents SDK logo
divider

Introduction

This guide walks you through connecting Firecrawl to the OpenAI Agents SDK using the Composio tool router. By the end, you'll have a working Firecrawl agent that can extract all product prices from this e-commerce site, crawl competitor blogs for latest article summaries, map all subpages linked from homepage url through natural language commands.

This guide will help you understand how to give your OpenAI Agents SDK agent real control over a Firecrawl account through Composio's Firecrawl MCP server.

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

Also integrate Firecrawl with

TL;DR

Here's what you'll learn:
  • Get and set up your OpenAI and Composio API keys
  • Install the necessary dependencies
  • Initialize Composio and create a Tool Router session for Firecrawl
  • Configure an AI agent that can use Firecrawl as a tool
  • Run a live chat session where you can ask the agent to perform Firecrawl operations

What is OpenAI Agents SDK?

The OpenAI Agents SDK is a lightweight framework for building AI agents that can use tools and maintain conversation state. It provides a simple interface for creating agents with hosted MCP tool support.

Key features include:

  • Hosted MCP Tools: Connect to external services through hosted MCP endpoints
  • SQLite Sessions: Persist conversation history across interactions
  • Simple API: Clean interface with Agent, Runner, and tool configuration
  • Streaming Support: Real-time response streaming for interactive applications

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

The Firecrawl MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Firecrawl account. It provides structured and secure access to automated web crawling, scraping, and data extraction, so your agent can perform actions like indexing sites, extracting structured content, mapping URLs, and searching the web on your behalf.

  • Automated web crawling and indexing: Let your agent launch and manage web crawl jobs to gather content or index entire websites efficiently.
  • Structured data extraction: Instruct your agent to extract targeted data from web pages using custom prompts or schemas, turning unstructured sites into actionable information.
  • URL mapping and discovery: Have the agent explore and map all URLs within a website, including options for subdomain inclusion, sitemap processing, or search-based discovery.
  • On-demand scraping and content retrieval: Enable your agent to scrape specific URLs, retrieve page content, and even extract structured JSON using LLM-powered methods.
  • Integrated web search and data collection: Task your agent with running web searches, scraping top result pages, and returning relevant details—all in one workflow.

Supported Tools & Triggers

Tools
Cancel an agent jobTool to cancel an in-progress agent job by its ID.
Batch scrape multiple URLsTool to scrape multiple URLs in batch with concurrent processing.
Cancel a batch scrape jobTool to cancel a running batch scrape job using its unique identifier.
Get batch scrape statusRetrieves the current status and results of a batch scrape job using the job ID.
Get errors from batch scrape jobTool to retrieve error details from a batch scrape job, including failed URLs and URLs blocked by robots.
Start a web crawlInitiates a Firecrawl web crawl from a given URL, applying various filtering and content extraction rules, and polls until the job is complete; ensure the URL is accessible and any regex patterns for paths are valid.
Cancel a crawl jobCancels an active or queued web crawl job using its ID; attempting to cancel completed, failed, or previously canceled jobs will not change their state.
Cancel a crawl jobTool to cancel a running crawl job by its ID.
Get crawl job statusTool to retrieve the status and results of a Firecrawl crawl job.
Get errors from a crawl jobTool to retrieve errors from a Firecrawl crawl job.
Get all active crawl jobsTool to retrieve all active crawl jobs for the authenticated team.
Preview crawl parametersPreview crawl parameters before starting a crawl by generating optimal configuration from natural language instructions.
Start a web crawl (v2) [NEW][NEW v2 API] Initiates a Firecrawl v2 web crawl with enhanced features over v1: natural language prompts for automatic crawler configuration, crawlEntireDomain for sibling/parent page discovery, better depth control with maxDiscoveryDepth, subdomain support, and full webhook configuration.
Get team credit usageTool to get current team credit usage information.
Get historical team credit usageTool to retrieve historical team credit usage on a monthly basis.
Extract structured dataExtracts structured data from web pages by initiating an extraction job and polling for completion; requires a natural language `prompt` or a JSON `schema` (one must be provided).
Get extract job statusTool to retrieve the status and results of a previously submitted extract job.
Get agent job statusTool to get the status and results of an agent job.
Get deep research statusRetrieves the status and results of a deep research job by its ID.
Get the status of a crawl jobRetrieves the current status, progress, and details of a web crawl job, using the job ID obtained when the crawl was initiated.
Generate LLMs.txt for a websiteInitiates an async job to generate an LLMs.
Get LLMs.txt generation job statusTool to get the status and results of an LLMs.
Map multiple URLsMaps a website by discovering URLs from a starting base URL, with options to customize the crawl via search query, subdomain inclusion, sitemap handling, and result limits; search effectiveness is site-dependent.
Get team queue statusTool to retrieve metrics about the team's scrape queue.
Scrape URLScrapes a publicly accessible URL, optionally performing pre-scrape browser actions or extracting structured JSON using an LLM, to retrieve content in specified formats.
SearchPerforms a web search for a query, scrapes content from the top search results using Firecrawl, and returns details in specified formats.
Start an agent jobTool to start an agent job for agentic web extraction with multi-page navigation and interaction capabilities.
Get team token usageTool to retrieve the current team's token usage and balance information for Firecrawl's Extract feature.
Get historical team token usageTool to retrieve historical team token usage on a monthly basis.

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

Prerequisites

Before starting, make sure you have:
  • Composio API Key and OpenAI API Key
  • Primary know-how of OpenAI Agents SDK
  • A live Firecrawl project
  • Some knowledge of Python or Typescript

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

Install dependencies

pip install composio_openai_agents openai-agents python-dotenv

Install the Composio SDK and the OpenAI Agents SDK.

Set up environment variables

bash
OPENAI_API_KEY=sk-...your-api-key
COMPOSIO_API_KEY=your-api-key
USER_ID=composio_user@gmail.com

Create a .env file and add your OpenAI and Composio API keys.

Import dependencies

import asyncio
import os
from dotenv import load_dotenv

from composio import Composio
from composio_openai_agents import OpenAIAgentsProvider
from agents import Agent, Runner, HostedMCPTool, SQLiteSession
What's happening:
  • You're importing all necessary libraries.
  • The Composio and OpenAIAgentsProvider classes are imported to connect your OpenAI agent to Composio tools like Firecrawl.

Set up the Composio instance

load_dotenv()

api_key = os.getenv("COMPOSIO_API_KEY")
user_id = os.getenv("USER_ID")

if not api_key:
    raise RuntimeError("COMPOSIO_API_KEY is not set. Create a .env file with COMPOSIO_API_KEY=your_key")

# Initialize Composio
composio = Composio(api_key=api_key, provider=OpenAIAgentsProvider())
What's happening:
  • load_dotenv() loads your .env file so OPENAI_API_KEY and COMPOSIO_API_KEY are available as environment variables.
  • Creating a Composio instance using the API Key and OpenAIAgentsProvider class.

Create a Tool Router session

# Create a Firecrawl Tool Router session
session = composio.create(
    user_id=user_id,
    toolkits=["firecrawl"]
)

mcp_url = session.mcp.url

What is happening:

  • You give the Tool Router the user id and the toolkits you want available. Here, it is only firecrawl.
  • The router checks the user's Firecrawl connection and prepares the MCP endpoint.
  • The returned session.mcp.url is the MCP URL that your agent will use to access Firecrawl.
  • This approach keeps things lightweight and lets the agent request Firecrawl tools only when needed during the conversation.

Configure the agent

# Configure agent with MCP tool
agent = Agent(
    name="Assistant",
    model="gpt-5",
    instructions=(
        "You are a helpful assistant that can access Firecrawl. "
        "Help users perform Firecrawl operations through natural language."
    ),
    tools=[
        HostedMCPTool(
            tool_config={
                "type": "mcp",
                "server_label": "tool_router",
                "server_url": mcp_url,
                "headers": {"x-api-key": api_key},
                "require_approval": "never",
            }
        )
    ],
)
What's happening:
  • We're creating an Agent instance with a name, model (gpt-5), and clear instructions about its purpose.
  • The agent's instructions tell it that it can access Firecrawl and help with queries, inserts, updates, authentication, and fetching database information.
  • The tools array includes a HostedMCPTool that connects to the MCP server URL we created earlier.
  • The headers dict includes the Composio API key for secure authentication with the MCP server.
  • require_approval: 'never' means the agent can execute Firecrawl operations without asking for permission each time, making interactions smoother.

Start chat loop and handle conversation

print("\nComposio Tool Router session created.")

chat_session = SQLiteSession("conversation_openai_toolrouter")

print("\nChat started. Type your requests below.")
print("Commands: 'exit', 'quit', or 'q' to end\n")

async def main():
    try:
        result = await Runner.run(
            agent,
            "What can you help me with?",
            session=chat_session
        )
        print(f"Assistant: {result.final_output}\n")
    except Exception as e:
        print(f"Error: {e}\n")

    while True:
        user_input = input("You: ").strip()
        if user_input.lower() in {"exit", "quit", "q"}:
            print("Goodbye!")
            break

        result = await Runner.run(
            agent,
            user_input,
            session=chat_session
        )
        print(f"Assistant: {result.final_output}\n")

asyncio.run(main())
What's happening:
  • The program prints a session URL that you visit to authorize Firecrawl.
  • After authorization, the chat begins.
  • Each message you type is processed by the agent using Runner.run().
  • The responses are printed to the console, and conversations are saved locally using SQLite.
  • Typing exit, quit, or q cleanly ends the chat.

Complete Code

Here's the complete code to get you started with Firecrawl and OpenAI Agents SDK:

import asyncio
import os
from dotenv import load_dotenv

from composio import Composio
from composio_openai_agents import OpenAIAgentsProvider
from agents import Agent, Runner, HostedMCPTool, SQLiteSession

load_dotenv()

api_key = os.getenv("COMPOSIO_API_KEY")
user_id = os.getenv("USER_ID")

if not api_key:
    raise RuntimeError("COMPOSIO_API_KEY is not set. Create a .env file with COMPOSIO_API_KEY=your_key")

# Initialize Composio
composio = Composio(api_key=api_key, provider=OpenAIAgentsProvider())

# Create Tool Router session
session = composio.create(
    user_id=user_id,
    toolkits=["firecrawl"]
)
mcp_url = session.mcp.url

# Configure agent with MCP tool
agent = Agent(
    name="Assistant",
    model="gpt-5",
    instructions=(
        "You are a helpful assistant that can access Firecrawl. "
        "Help users perform Firecrawl operations through natural language."
    ),
    tools=[
        HostedMCPTool(
            tool_config={
                "type": "mcp",
                "server_label": "tool_router",
                "server_url": mcp_url,
                "headers": {"x-api-key": api_key},
                "require_approval": "never",
            }
        )
    ],
)

print("\nComposio Tool Router session created.")

chat_session = SQLiteSession("conversation_openai_toolrouter")

print("\nChat started. Type your requests below.")
print("Commands: 'exit', 'quit', or 'q' to end\n")

async def main():
    try:
        result = await Runner.run(
            agent,
            "What can you help me with?",
            session=chat_session
        )
        print(f"Assistant: {result.final_output}\n")
    except Exception as e:
        print(f"Error: {e}\n")

    while True:
        user_input = input("You: ").strip()
        if user_input.lower() in {"exit", "quit", "q"}:
            print("Goodbye!")
            break

        result = await Runner.run(
            agent,
            user_input,
            session=chat_session
        )
        print(f"Assistant: {result.final_output}\n")

asyncio.run(main())

Conclusion

This was a starter code for integrating Firecrawl MCP with OpenAI Agents SDK to build a functional AI agent that can interact with Firecrawl.

Key features:

  • Hosted MCP tool integration through Composio's Tool Router
  • SQLite session persistence for conversation history
  • Simple async chat loop for interactive testing
You can extend this by adding more toolkits, implementing custom business logic, or building a web interface around the agent.

How to build Firecrawl MCP Agent with another framework

ChatGPT logo

ChatGPT

Use Firecrawl MCP with ChatGPT

Claude Agent SDK logo

Claude Agent SDK

Use Firecrawl MCP with Claude Agent SDK

Claude Code logo

Claude Code

Use Firecrawl MCP with Claude Code

Claude Cowork logo

Claude Cowork

Use Firecrawl MCP with Claude Cowork

Codex logo

Codex

Use Firecrawl MCP with Codex

Cursor logo

Cursor

Use Firecrawl MCP with Cursor

OpenClaw logo

OpenClaw

Use Firecrawl MCP with OpenClaw

Hermes logo

Hermes

Use Firecrawl MCP with Hermes

CLI logo

CLI

Use Firecrawl MCP with CLI

Google ADK logo

Google ADK

Use Firecrawl MCP with Google ADK

LangChain logo

LangChain

Use Firecrawl MCP with LangChain

Vercel AI SDK logo

Vercel AI SDK

Use Firecrawl MCP with Vercel AI SDK

Mastra AI logo

Mastra AI

Use Firecrawl MCP with Mastra AI

LlamaIndex logo

LlamaIndex

Use Firecrawl MCP with LlamaIndex

CrewAI logo

CrewAI

Use Firecrawl MCP with CrewAI

OpenCode logo

OpenCode

Use Firecrawl MCP with OpenCode

VS Code logo

VS Code

Use Firecrawl MCP with VS Code

FAQ

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

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

Can I use Tool Router MCP with OpenAI Agents SDK?

Yes, you can. OpenAI Agents SDK 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 Firecrawl tools.

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

Yes, absolutely. You can configure which Firecrawl 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 Firecrawl data and credentials are handled as safely as possible.

Used by agents from

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

Get started for freeGet a demo↗