fatsearch
FATSEARCH is an open-source command-line tool and local MCP server for searching and reading web pages for AI agents. It aims to replace large, noisy page fetches with short, cached Markdown outlines that agents can read section by section, using no account or personal telemetry on the website.
At a Glance
- AI-agent and LLM application developers
- Software engineers using Claude Code, Codex, Cursor, or other MCP-capable agents
- Developers needing local CLI web extraction
- Researchers and automation builders handling blocked, JavaScript-heavy, or high-context web pages
AI Tools by fatsearch
(1)
FATSEARCH
Web Reader MCP Server for Agents
Discussions
No discussions yet
Be the first to start a discussion about fatsearch
Latest News
FATSEARCH 0.2.5 released with MCP setup for Claude Code, Codex, and Cursor, provider fallback/security fixes, and cross-platform binaries.
FATSEARCH 0.2.4 released with improved provider setup, MCP host detection, cache reuse, search behavior, and agent guidance.
FATSEARCH 0.2.3 added visible waits/progress for long fetches and improved search, post counts, rendering, and agent guidance.
FATSEARCH 0.2.2 added Windows installation through npm and fixed platform, cache, and key-handling behavior.
Products & Services
A Rust command-line web reader and search tool. It fetches pages and documents, produces compact Markdown outlines with section IDs, supports targeted section reads and filters, caches results locally, and can use a free direct request or user-supplied provider keys for harder pages.
A local stdio Model Context Protocol server exposing one fats tool so coding agents can invoke the same outline, search, filter, and section-reading workflow.
A versioned skill in the GitHub repository that teaches agents how to install and use fats, including when to search, how to read cached pages, and how to configure providers.
Market Position
FATSEARCH positions itself as the 'ripgrep of web searching for AI agents': a lightweight, local, outline-first reader rather than a hosted crawler that returns an entire page. Its own benchmarks compare it with Firecrawl, Tavily, Jina Reader, Crawl4AI, SearXNG-based MCP tools, and other readers, emphasizing substantially lower context/token use and free plain-page fetching, while acknowledging that Firecrawl is better suited to whole-site crawling and batch extraction and that hosted search readers can be faster or simpler for some jobs.
Leadership
Founders
guyposka
GitHub maintainer and author of the FATSEARCH repository; the GitHub history attributes the first public release and subsequent release work to this account. The profile also lists the public projects sub-500-ai-voice and zero-fail-scrape-fleet.
Executive Team
guyposka
Creator and maintainer
GitHub account behind the fatsearch organization and the repository's release history; credited with the first public release and later FATSEARCH releases.
Founding Story
FATSEARCH was started as an open-source answer to the problem that AI agents' built-in web fetchers either cannot read some pages or return badly shaped, overly large context. Its initial vision was a small, local, provider-key-compatible reader that returns an outline first and lets an agent open only the relevant section.
Business Model
Revenue Model
FATSEARCH itself is free and open source. Users bring their own provider credentials; any usage charges are paid to the selected web-data/search provider rather than through a FATSEARCH subscription or hosted account.
Pricing Tiers
Direct page fetching works without a key for pages accessible from the user's connection.
The project's comparison page reports this approximate provider-cost range for pages requiring Oxylabs or Bright Data; the key and provider billing are the user's responsibility.
Target Markets
- AI-agent and LLM application developers
- Software engineers using Claude Code, Codex, Cursor, or other MCP-capable agents
- Developers needing local CLI web extraction
- Researchers and automation builders handling blocked, JavaScript-heavy, or high-context web pages
- Giving coding and research AI agents compact, relevant web context
- Reading pages that built-in agent fetchers cannot access or parse well
- Reducing token usage and model-context cost when extracting facts from long pages
- Searching the web and then opening only the relevant result sections
- Local, privacy-conscious web reading without an account or personal site telemetry