Firecrawl MCP Server

An MCP server for Firecrawl - web scraping and search with batch support, rate limiting, and flexible output modes.

Release status (2026-07-27): this repository intentionally pins @modelcontextprotocol/server and the test-only @modelcontextprotocol/client to 2.0.0-beta.5. These packages implement the 2026-07-28 release candidate, not a final published protocol. Re-run the final release gate before claiming final conformance.

The server uses the public serveStdio entry with legacy mode set to serve. The SDK chooses the protocol era from the opening exchange and pins one server instance to that stdio connection. Standard output is reserved for MCP frames; diagnostics are written only to standard error.

Features

• Scrape Tool: Extract content from web pages

  • Single URL or batch scraping (up to 100 URLs)

  • Multiple output formats: markdown, HTML, links, screenshot

  • Smart content extraction (main content only)

  • Geographic targeting

• Search Tool: Search the web

  • Single or batch queries (up to 10)

  • Web, image, and news results

  • Location-based search

  • Time filtering (hour/day/week/month/year)

  • Optional content scraping from results

• Rate Limit Handling: Automatic retry with exponential backoff

• Two Output Modes:

  • direct: Return results to the agent

  • file: Save as markdown files with date-based structure

Related MCP server: DuckDuckGo MCP Server

Quick Start

# Install dependencies
bun install

# Set your Firecrawl API key
export FIRECRAWL_API_KEY=fc-your-api-key

# Run in development mode
bun run dev

# Test with MCP Inspector
bun run inspector

Configuration

All configuration via environment variables:

Tools

scrape

Scrape web pages and extract content.

{
  "urls": "https://example.com",
  "formats": ["markdown"],
  "onlyMainContent": true,
  "outputMode": "direct"
}

Parameters:

• urls: Single URL or array of URLs (max 100)

• formats: Output formats - markdown, html, rawHtml, links, screenshot

• onlyMainContent: Extract only main content (default: true)

• includeTags: HTML tags to include

• excludeTags: HTML tags to exclude

• waitFor: Wait time in ms for dynamic content

• timeout: Request timeout in ms

• location: Geographic targeting { country: "US", languages: ["en"] }

• outputMode: Override mode - direct or file

• outputDir: Override output directory

search

Search the web and optionally scrape results.

{
  "queries": "firecrawl web scraping",
  "limit": 10,
  "scrapeResults": true,
  "outputMode": "direct"
}

Parameters:

• queries: Single query or array of queries (max 10)

• limit: Results per query (max 20, default: 10)

• location: Search location (e.g., "Germany", "United States")

• tbs: Time filter - qdr:h (hour), qdr:d (day), qdr:w (week), qdr:m (month), qdr:y (year)

• scrapeResults: Also scrape content from results

• scrapeFormats: Formats for scraped content

• outputMode: Override mode - direct or file

• outputDir: Override output directory

Output Modes

Direct Mode (default)

Results are returned directly to the agent in the response.

File Mode

Results are saved as markdown files:

firecrawl-output/
├── scrape/
│   └── 2026-01-25/
│       └── example.com/
│           ├── index.md
│           └── about.md
└── search/
    └── 2026-01-25/
        └── 14-30-00/
            └── firecrawl-web-scraping.md

Each file includes YAML frontmatter with metadata:

---
url: https://example.com
title: Example Domain
scraped_at: 2026-01-25T14:30:00.000Z
---

# Example Domain

This domain is for use in illustrative examples...

Client Configuration

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "firecrawl": {
      "command": "bun",
      "args": ["run", "/path/to/firecrawl-mcp/src/index.ts"],
      "env": {
        "FIRECRAWL_API_KEY": "fc-your-api-key",
        "FIRECRAWL_OUTPUT_MODE": "direct"
      }
    }
  }
}

Cursor

Add to MCP settings:

{
  "firecrawl": {
    "command": "bun",
    "args": ["run", "/path/to/firecrawl-mcp/src/index.ts"],
    "env": {
      "FIRECRAWL_API_KEY": "fc-your-api-key"
    }
  }
}

Rate Limiting

The server includes built-in rate limiting to handle Firecrawl's API limits:

• Token bucket algorithm for request throttling

• Automatic retry on 429 (rate limit) and 5xx errors

• Exponential backoff between retries

• Configurable limits via environment variables

Development

# Lint
bun run lint

# Type check
bun run typecheck

# Format code
bun run format

# Run the official-client source protocol matrix
bun run test:protocol

# Build for distribution and repeat the handshake against Node
bun run build
bun run test:protocol:built

# Check formatting
bun run format:check

License

MIT