firecrawl-lite

Firecrawl Lite is a private, standalone MCP server that provides web scraping and data extraction capabilities using your own local LLM setup. It runs entirely on your machine, with no need for external API calls beyond your chosen LLM provider. The server exposes a set of tools for page scraping, batch processing, and data extraction, allowing you to fetch content, analyze pages with prompts, and structure results according to schemas. Tools include: scrape_page for single-page content extraction, batch_scrape for processing multiple URLs at once, extract_data for natural-language driven data extraction, extract_with_schema for schema-driven extraction, and screenshot to capture page visuals. You can configure your LLM provider, API key, and model to tailor the behavior to your environment, whether you’re using OpenAI, xAI, Anthropic, Ollama, or a local LLM.

To use the tools, start the MCP server (via npx as shown in the configuration), then send JSON-RPC style requests or use your MCP-integrated clients (Claude, Cursor, etc.) to invoke the available methods. For example, to scrape content from a page, call the scrape_page tool with the URL; to extract structured data with a prompt, use extract_data; to fetch a page and capture its main content or a schema-defined object, use extract_with_schema; and to capture a visual screenshot, use the screenshot tool with width, height, and fullPage options. The server supports remote deployment via Docker or remote HTTP endpoints, and includes guidance for enabling HTTP streaming and SSE endpoints as needed.

If you integrate with Claude, Cursor, or other MCP clients, add the server as an MCP source using the specified command and environment variables so your client can communicate with the local Firecrawl Lite instance.

Prerequisites:

• Node.js (for using npx to install the MCP server)
• Internet access to fetch the MCP server package
• A local LLM setup or-api-access for your chosen provider (keys and base URL)

Installation steps:

1. Ensure Node.js and npm are installed on your system.

   • Check: node -v and npm -v

2. Install and run the MCP server via npx using the recommended configuration:

   • Create a short run/config file if desired, but you can directly run the command below:

   npx -y @ariangibson/firecrawl-lite-mcp-server

3. Configure environment variables (LLM credentials and model):

   • LLM_API_KEY=your_api_key_here
   • LLM_PROVIDER_BASE_URL=https://api.x.ai/v1 (or your provider base URL)
   • LLM_MODEL=grok-code-fast-1 (or your preferred model)

4. If embedding in an editor or client, export the configuration as shown in the mcp_config example and reuse the same env settings.

5. Optional: For remote deployment, consider Docker deployment or enabling HTTP endpoints as described in the README.

Note: If you are using a package manager or a GUI-driven integration, ensure your MCP client is configured to point to the local MCP server (e.g., http://localhost:3000/mcp or equivalent) and that any required environment variables are passed to the client as part of the integration.

Tips and notes:

• Privacy-first architecture ensures all web scraping and data extraction runs locally; ensure your LLM provider configuration aligns with your local setup.
• If you encounter Chrome/puppeteer issues, you can manually install Chromium or reset Puppeteer caches as described in the troubleshooting section of the README.
• When deploying remotely (Docker or HTTP endpoints), enable at least one endpoint to support the desired transport (HTTP streaming or SSE).
• For anti-detection adjustments, you can configure user agents and scrape delays to mimic natural browsing behavior.
• If using Claude or other MCP clients, you can copy the exact mcpServers configuration snippet from the README into your client configuration to ensure parity across tools.
• The official images are available on Docker Hub and GitHub Container Registry for convenient deployment across architectures (amd64, arm64).