Connect Firecrawl to an MCP client with OAuth, an API key, or a keyless trial.
Firecrawl MCP gives AI clients access to live web data through the Model Context Protocol. Use the hosted server for the fastest setup, or run the open-source server locally when you need a local process or self-hosted Firecrawl API.This page is the canonical setup guide for hosted Firecrawl MCP.
For an interactive client, connect with OAuth. The client opens Firecrawl in your browser so you can sign in, choose a team, and approve access without copying an API key.
Save the file, open Cursor’s MCP settings, and complete the Firecrawl sign-in. Cursor then makes the Firecrawl tools available to Agent.
Point an OAuth-capable remote MCP client at:
https://mcp.firecrawl.dev/v2/mcp-oauth
If the client asks for an OAuth Client ID or Client Secret, leave both blank. Compatible clients use Client ID Metadata Documents or Dynamic Client Registration.If the client cannot complete remote OAuth, use the API-key setup below. If it cannot send request headers, run Firecrawl MCP locally.
OAuth connections receive short-lived, resource-bound tokens rather than your Firecrawl API key. You can review and revoke them from MCP settings.
Use an API key for CI, servers, scripts, or another client that cannot complete browser OAuth. Create a Firecrawl API key, keep it in an environment variable or secret store, and connect to the full hosted MCP endpoint.
Header configuration differs by client. Keep the key in the client’s environment or protected secret storage.
Never put an API key in a project configuration file, paste it into chat, or share it in a screenshot or issue. New configurations should use OAuth or an environment-backed Authorization header, never a credential-bearing URL.
To try Firecrawl without creating an account, add the keyless endpoint to any client that supports remote MCP:
https://mcp.firecrawl.dev/v2/mcp
Keyless MCP is rate-limited per network and exposes exactly Search, Scrape, and Parse while eligible. Connect with OAuth or an API key when you need the full tool surface or higher limits.
What happens when I reach the keyless limit?
The MCP error includes the current retry time when it is available. Wait for that interval to continue keyless use, or connect immediately through OAuth or an API key.
Open your client’s MCP status or tool list and confirm that firecrawl is connected. A keyless connection shows firecrawl_search, firecrawl_scrape, and firecrawl_parse; an authenticated connection can expose the full tool surface, subject to plan, deployment, and team policy.Try a read-only request:
Search the web for the latest Firecrawl release notes and summarize the sources.
If no Firecrawl tools appear, restart or reload the client after saving its MCP configuration. For a local-process error such as spawn npx ENOENT, see Run Firecrawl MCP locally.
This legacy form exists only for existing configurations and clients that cannot complete OAuth or send a custom Authorization header. It is not the recommended setup for new integrations.
The API key is part of the URL, so treat the complete URL like a password. If it is exposed, rotate the key. This fallback applies only to the full /v2/mcp surface. It does not work for the OAuth-only search resource.
Existing OAuth connections
Existing /v2/mcp OAuth connections remain supported during the migration to the dedicated account endpoint. New interactive account connections should use /v2/mcp-oauth; new tokens for that account resource are not accepted by /v2/mcp.
