fastCRW MCP Server: Scrape, Crawl and Search the Web with AI

• Turns one web page into markdown

  Fetches one URL and returns it as markdown, HTML or a list of links. Output is cut to about 15,000 characters unless you ask for all of it.

  Try: “Scrape https://example.com/pricing as markdown”

  Network · crw_scrape

• Starts a crawl of a whole site

  Starts a background crawl from a start URL and returns a job ID. You poll the status tool for the pages.

  Try: “Crawl docs.example.com up to 50 pages”

  Write · crw_crawl

• Collects the pages from a crawl job

  Checks a crawl job by ID and returns its status and the pages gathered so far.

  Try: “Is crawl job abc123 finished? Show me the pages”

  Network · crw_check_crawl_status

• Lists the URLs on a site

  Finds URLs through the sitemap and a short crawl and returns only the list, not page content. It returns at most 100 URLs unless you lift the limit.

  Try: “Map all the URLs on example.com”

  Network · crw_map

• Pulls structured fields from pages

  Reads one or more URLs and returns JSON built from your prompt, a JSON schema or both. It runs as a background job and needs an LLM configured for CRW.

  Try: “Extract product name and price from these 5 URLs”

  Write · crw_extract

• Gets the result of an extract job

  Polls an extract job by ID and, when it is done, returns results for each URL.

  Try: “Check extract job xyz789”

  Network · crw_check_extract_status

• Cancels a running extract job

  Asks CRW to stop an extract job and returns its final status. Cancelling can take a moment while the URL being read finishes.

  Try: “Cancel extract job xyz789”

  Write · crw_cancel_extract

• Searches the web

  Runs a web search and returns ranked results with title, URL and snippet, and can scrape each result. In local mode it only appears when you have set up a search backend. With a fastcrw.com key it is always there.

  Try: “Search for the latest Rust async runtime benchmarks”

  Network · crw_search

• Turns a PDF into markdown

  Takes a PDF you pass as base64 and returns markdown. There is no OCR, so scanned PDFs come back empty with a warning.

  Try: “Convert this PDF to markdown”

  Read · crw_parse_file

Watch out for

• Web pages can carry instructions aimed at the assistant

  Scraped text goes straight into the conversation, and crw_search tells the assistant to use it for anything that may need fresh facts. Our advice: review what the assistant does after reading pages you do not control. github.com

• The install command edits your AI tools' configs

  The npx install command adds a skill and registers the server in every detected tool, such as Claude Code, Cursor and Codex, and keeps a .crw-backup copy of each config it changes. Use the install box instead if you want one tool only. npmjs.com

• It may download a program on first run

  The npm package normally carries a prebuilt binary. If that is missing for your system, the launcher downloads one from the project's GitHub release and checks it against the published checksums. npmjs.com

• Two fixed advisories in the CRW engine

  Both were fixed long before the current release, so no action is needed on 0.39.1. Update if you run an older self-hosted crw-server. github.com

Security advisories