Ransack - Web Search & Fetch MCP Server for AI Agents

Ransack is an MCP server that gives AI agents live web search and page fetch through one endpoint and one API key. Twelve search engines behind a single MCP connection: search, fetch, discover, hosts, shop, verify, plus YouTube. No credits to manage, no per-engine billing: $15/month flat, 500 calls/day, failed fetches cost nothing, and every fetch returns a pass/fail receipt so silent failures cannot burn you.

This is a hosted service. This repository contains connection configuration, documentation, and logo; the server source is not public. Connect to:

   https://ransack.tools/mcp

• Homepage: ransack.tools
• Free trial key: ransack.tools/trial

Connect

Get an API key from ransack.tools/trial, then add the server to your client.

Claude Desktop

{"mcpServers":{"ransack":{"type":"streamable-http","url":"https://ransack.tools/mcp","headers":{"Authorization":"Bearer <API_KEY>"}}}}

Cursor

{"mcpServers":{"ransack":{"url":"https://ransack.tools/mcp","headers":{"Authorization":"Bearer <API_KEY>"}}}}

Pi / Cline / 5ire

{"mcpServers":{"ransack":{"type":"http","url":"https://ransack.tools/mcp","headers":{"Authorization":"Bearer <API_KEY>"}}}}

Transport is streamable-http; authentication is a Bearer API key.

Tools

ransack

One tool, many modes: search, fetch, discover, hosts, shop, verify, research. search is metasearch across engines; fetch is the primary answer path; hosts enumerates a domain's assets from certificate transparency, DNS, sitemaps, web archives, crawl indexes, JS routes and scan records; shop returns price and availability; verify checks a claim against its cited sources. freshness bypasses the result cache; it does not date-filter, so results are not filtered by date.

youtube

Fetch a YouTube transcript from a video URL.

execute_research

Multi-step pipeline (discover, fetch, extract, synthesize) into a cited report. Returns a task handle by default; poll tasks_get for the report_id and its per-claim citation verdicts.

get_report

Retrieve a persisted report by report_id.

search_memory

Semantic recall over previously fetched page chunks.

tasks_get

Poll an asynchronous research task.

What you actually get back

• Engine-tagged results. Every search result names the engine that found it, so you can tell Google from an independent index.
• Coverage status, not silence. On hosts, every discovery engine reports ok, empty, timeout or skipped. Found-nothing is never indistinguishable from tried-and-failed.
• Labelled walls. [bot wall: <vendor>] and [dead page: HTTP 404] come back as labels. Pages the server cannot read are reported, not guessed.
• Per-claim citation verdicts. Research reports carry a verdict per claim, checked against the sources actually retrieved.
• No query rewrites. Ever. The caller owns site selection. Ransack never appends operators to your query or silently changes what you asked for.
• One key, six tools. Search, fetch, discovery, shopping, transcripts and cited research behind a single endpoint.

Honest limitations

• Paywalled and login-walled pages (e.g. some news, GitHub /commits views) can't always be read.
• Retailers that hard-block scraping (Amazon, Walmart) may return nothing rather than guessed data.
• TikTok is searchable, and the result snippet already carries the caption. The reader tier returns the caption, creator handle, post date and engagement counts, but the video, the comments and the spoken transcript do not come back, and the browser tier can return only TikTok's "Page not available" stub.
• search fast-lane results are partial by design; fetch is the primary answer path.
• Labelling is not guaranteed: a wall sometimes comes back as thin page content instead.

Pricing

• Trial: free for 14 days, 50 calls/day, no card (ransack.tools/trial).
• Paid: $15/month flat, 500 calls/day, 120 requests/min.

License

This repository is documentation and configuration only. The service at ransack.tools is proprietary.