free-search-mcp
A local MCP server with no API keys: multi-engine web search, page fetching and document reading tuned for LLMs.
- 177 ★ on the repository
- Last commit
Extended access
Why this level
- The server fetches any address the agent asks for and stores pages in a local cache.
- The HTTP mode has no authentication: an open port lets anyone who reaches it fetch URLs.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/free-search-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server fetches any address the agent asks for and stores pages in a local cache.
- The HTTP mode has no authentication: an open port lets anyone who reaches it fetch URLs.
This is third-party code. Review the repository files before installing.
What it does
free-search-mcp runs on your machine and gives any model web search, page fetching and document reading with no search API sign-up. A query goes to several engines in parallel (DuckDuckGo, Mojeek, Startpage and others), results are merged and deduplicated. Pages are converted to Markdown with a token estimate and errors name the next step. For JavaScript-heavy pages there is a Playwright and Chromium fallback. There are 11 tools: `search` and `research` (search plus fetching the top sources in one call), `compare` for two to five links, `fetch` and `fetch_batch` (up to 20 addresses), `read_doc` for PDF, DOCX, XLSX, PPTX, EPUB, CSV and code with paged reading, `extract_structured` for JSON-LD and OpenGraph, `cache_search` over the FTS5 full-text cache, `paper_graph` for a paper citations, `engines` and `download`. Filters cover freshness, domains and categories (news, papers, software, security). The plugin adds the verified-research skill and the quick-search subagent for Claude Code.
Who it is for. People who want to give Claude Code, Codex or a local model web search and page reading without paid search APIs or sign-ups.
Good fit when
- The agent needs a reference for a fresh library: `research` assembles a short brief with sources.
- Compare the terms of two services across five links and get excerpts for a question.
- Read a long PDF page by page and find the section needed.
- Find a paper and walk through its references and citations with `paper_graph`.
Not a fit when
- Stable search results with a contract and SLA are needed: these are free engines with no guarantees.
- The server must be exposed to a network: the HTTP mode has no authentication and fetches any URL for whoever reaches it.
- Pages behind login or bot protection: the server detects CAPTCHA and reports an error but does not solve it.
Example request
Using free-search, study the new pagination rules in the library docs, assemble a short brief with sources and show which pages were read.MCP
- Transport
- stdio, http
- Authentication
- not required
Environment variables
SEARCH_MCP_CACHE_DIR- Cache and downloads directory
SEARCH_MCP_DOWNLOAD_ENABLED- false disables the download tool
Limitations
Python 3.11 or newer and uv are required. Some engines (`brave`, `zhihu`) and JavaScript-heavy pages need Chromium, installed once with `uvx --from free-search-mcp playwright install chromium`. The HTTP mode (`--transport streamable-http`) has no authentication and should stay on loopback or behind an authenticating proxy. The `download` tool saves files to `SEARCH_MCP_CACHE_DIR/downloads` and deletes them after a day, disabled with `SEARCH_MCP_DOWNLOAD_ENABLED=false`. A twelfth tool, `ask`, appears only when an answer backend is configured.
How to disable. For Claude Code run `claude mcp remove search` or `/plugin uninstall free-search`; for Codex `codex mcp remove search`. The cache lives in `SEARCH_MCP_CACHE_DIR` and can be deleted by hand.
FAQ
How do I install free-search in Claude Code?
With `/plugin marketplace add sweetcornna/free-search-mcp` and `/plugin install free-search@free-search-mcp`. Without the plugin, `claude mcp add search -s user -- uvx free-search-mcp` works.
Is an API key needed?
No, no search API keys are required. A key is needed only for the optional backend of the `ask` tool.
Which document formats does read_doc read?
PDF, DOCX, XLSX, PPTX, EPUB, CSV, zip and tar archives, HTML, TXT, Markdown and source code, with paged reading.
What does the server do with CAPTCHA?
An engine that serves a CAPTCHA or an empty page is skipped with a note in the error, there is no CAPTCHA solving.
Which search engines are queried?
DuckDuckGo, Mojeek, Startpage, Bing and others in the examples, the `engines` tool shows the list.
Discussion
No questions or reviews yet.
Related
MCP serversLocal web search for an agent with no API keys: results are ranked on your machine and only the relevant passages come back
Web search via DuckDuckGo and page reading for the agent with no API keys, with built-in rate limiting
A research MCP server: searches the web, SEC, courts and patents, reads whole pages and checks that cited sources really exist.
The official Search1API MCP: web search, news, page crawling, sitemaps and trends with OAuth sign-in or an API key.
Metasearch for agents: one MCP server and HTTP API merge results from many search providers into unified JSON with deduplication.
Local web search for agents: searches, reads pages and returns compact passages with sources, saving context tokens
README in short
The README explains the purpose of the server, a quick start through Claude Code and Codex plugins, a comparison with other search MCP servers, a table of 11 tools with filters, other install routes (Claude Desktop, uvx, Docker, installer script) and engine handling.