Firecrawl MCP Server

Firecrawl MCP Server is the official Model Context Protocol server from Firecrawl (firecrawl.dev, previously hosted under the GitHub name "mendableai"). It connects AI clients such as Claude Code, Cursor, VS Code, and Windsurf to Firecrawl's web crawling, search, and extraction services so an agent can retrieve current web content in a structured way instead of guessing. The direct product and documentation link is https://docs.firecrawl.dev/mcp-server, and the repository at https://github.com/firecrawl/firecrawl-mcp-server holds the source code, the MIT license, the version state, and concrete client configurations. For Skill Road this is a distinct entry because it does considerably more than a plain page fetch: it renders JavaScript, follows links in depth, returns clean Markdown or JSON, and bundles multi-source research into one service.

Tools and output formats

According to the README, the server exposes several clearly separated tools. firecrawl_scrape retrieves the content of a single URL and returns it as Markdown, as structured JSON, or with layout information. firecrawl_map discovers the reachable URLs of a website, and firecrawl_search combines web search with optional scraping of the results. firecrawl_crawl and firecrawl_check_crawl_status process many pages with depth and volume limits as an asynchronous job. firecrawl_parse turns files and documents such as PDF, DOCX, XLSX, or HTML into prepared Markdown. There is also firecrawl_agent for autonomous multi-source research, firecrawl_interact for clicking and typing on a page, a set of firecrawl_research_* tools for academic literature, firecrawl_developer_search for code-adjacent sources, and a firecrawl_monitor_* group for recurring page monitoring. Firecrawl explicitly advises against passing several URLs to a single scrape call; genuine bulk operations belong to the batch endpoint of the regular Firecrawl API outside of MCP.

Hosted, local, and self-operated

There are three paths. The hosted server runs at https://mcp.firecrawl.dev/v2/mcp and offers a keyless trial mode for compatible clients: according to the documentation, scrape, search, and parse work without an API key, though rate-limited. Tools such as crawl, map, and agent require a Firecrawl API key or a sign-in through the OAuth flow. Locally, the server starts with npx -y firecrawl-mcp and the FIRECRAWL_API_KEY environment variable; stdio, Server-Sent Events, and a streamable HTTP mode are supported. Teams that run Firecrawl themselves point FIRECRAWL_API_URL at their own instance. Automatic retries and rate limiting are built in according to the provider.

How it differs from Fetch MCP and Context7

Skill Road already lists Fetch MCP and Context7. Fetch MCP retrieves single pages and converts them to Markdown; Context7 provides version-accurate library documentation. Firecrawl MCP Server operates one level above: multi-page crawling, web search with extraction, JavaScript rendering, schema-based structured fields, and document parsing in one service. For a single static article, a lightweight fetch remains sensible; once whole documentation areas, directories, index pages, or recurring monitoring are needed, Firecrawl plays to its strengths.

Cost, license, and GitHub stars

The MCP server is MIT licensed; the GitHub star count was around 7,410 on 2026-09-07, and the current version is v3.2.1. That number is a point-in-time snapshot of popularity, not a security or quality guarantee. The keyless mode is free for first tests; beyond that, the Firecrawl plans apply with paid quotas and enterprise tiers, whose current terms are stated on Firecrawl's official pricing page. This public entry deliberately avoids fixed amounts. The Research, Coding, and Automation categories fit because the server connects source gathering, development work, and repeatable crawl jobs.

Who the server is for

Firecrawl MCP Server makes sense for teams that want to supply agents with verifiable, current web sources: building RAG, competitive and market monitoring, documentation research, or data collection across many subpages. Anyone who only occasionally fetches a known page can use a simpler tool. Before connecting it, teams should decide which target sites are allowed, respect the rate limits and terms of use of those sources, and keep the API key only in environment variables or secure client inputs.

MCP client, Node.js/npm for local operation, and depending on the tool a Firecrawl API key or OAuth sign-in.

Connect the hosted https://mcp.firecrawl.dev/v2/mcp (keyless trial for scrape/search/parse) or run local npx -y firecrawl-mcp with FIRECRAWL_API_KEY.

npx -y firecrawl-mcp

Keyless trial for scrape/search/parse; crawl/map/agent need a Firecrawl API key or OAuth sign-in in the client.

The server fetches web pages and documents on the agent’s behalf; scope follows the selected tool and the API key quota.

Target URLs, search terms, and uploaded files go to the Firecrawl service; prepared content returns to the MCP client.

Uncontrolled target inputs, API keys in prompts or config files, and crawling third-party sites without regard for their terms of use can put data and credentials at risk.