Log inSubmit server
Free Search logo

Free Search

Local-first, no-key MCP server for multi-engine search, web fetching, document parsing, caching, and source comparison.

InstallRepositoryWebsite
Full README Compare alternatives Embed widget Claim this listing

Install Free Search

Pick your client, then add it in one click or paste the config.

STDIOPython
~/Library/Application Support/Claude/claude_desktop_config.json
claude_desktop_config.json
{
  "mcpServers": {
    "free-search": {
      "command": "uvx",
      "args": [
        "free-search-mcp"
      ]
    }
  }
}

Settings → Developer → Edit Config, merge this into mcpServers, then fully quit and reopen Claude Desktop.

Install not yet sandbox-tested— the server itself is confirmed live (we called its tools/list endpoint). The stdio install command is a separate, ongoing check.

Overview

Free Search MCP server connects MCP clients to keyless web search, page retrieval, document parsing, and local cache search. Its 11 tools cover multi-engine discovery, one-shot research, citation graphs, structured metadata extraction, comparisons, downloads, and more. Results are Markdown by default or structured JSON, with caching and source metadata included in many workflows. Reach for it when an agent needs current web information without signing up for a search API.

Use cases

  • Search current technical documentation and software versions
  • Research a question and return source-backed page text
  • Read and paginate remote PDF or DOCX files
  • Compare several product or article URLs
  • Search previously fetched pages in the local cache

Key features

  • No-API-key multi-engine web search
  • Reader-mode page fetching and document parsing
  • One-call search and research briefs
  • Local SQLite FTS5 cache search
  • Paper citation and retraction graph lookup
  • Structured metadata extraction and temporary downloads

Tools11

What Free Search lets your agent do. Expand a tool for its parameters.

Verified live~7.9k tokens of context

Security & health

What it can reach and how reliably it runs — from our own checks, not the maintainer’s claims.

ActiveLast checked 1d ago

Recent health check succeeded.

15/15 checks healthy over the last 44d

What it can do

From the 11 tools this server returned when we ran it; 11 declare their own safety hints.

Handles credentials · 3Makes network requests · 10Read-only tools · 11
Network at startup2 connections
Files written at startup24

Observed 2026-10-24 in an isolated sandbox. A description of reach, not a security audit.

Supply-chain signal

No high-severity advisories surfaced by our automated scan.

Critical 0High 0Medium 0Low 0

Scanned 13d ago via OSV.dev · free-search-mcp (PyPI)

Adoption & maintenance

Factual signals from GitHub, npm, and our automated checks — not a rating.

GitHub stars176Stargazers on the source repository.
Last commit12d agoMost recent push to the default branch.
Tools exposed11Callable tools this server registers over MCP.
Directory activity8 viewsConfig copies, upvotes, and views on AllMCPs.

How Free Search works

What Free Search does

Free Search provides a local-first MCP interface for discovering and reading web content without requiring search API credentials. It combines several keyless search sources, merges and deduplicates their results, and can route queries toward source categories such as papers, finance, software, security, weather, datasets, and government information.

The server also handles content after discovery. HTML pages are cleaned into reader-style Markdown, common office and document formats are parsed into text, and non-text resources can be described by type, size, dimensions, and SHA-256 hash. Search and fetch results are cached locally, allowing later full-text queries over previously fetched pages.

How it works

A typical workflow starts with search, which accepts a natural-language query and returns ranked links. The default engine selection is keyless and can adapt when a source becomes unavailable. category, domain filters, text filters, and best-effort freshness windows refine the search. Explicit engine selection is available when a particular index is needed; engines lists the available source tree.

For an open-ended question, research combines discovery with fetching the top results and returns a citable brief. Known URLs should go to fetch, while multiple URLs can be handled concurrently with fetch_batch or compared with compare. Search and page data use a default seven-day cache, with refresh controls available for time-sensitive content.

The server runs locally through the MCP client and uses Python 3.11 or newer. Browser-based rendering is optional: Chromium can be installed for JavaScript-heavy pages and browser-dependent engines, while ordinary HTTP retrieval works without it.

Setup and configuration

The package can be launched with uv, so a source checkout is not required. Claude Code and Codex support the repository's plugin installation flow. Claude Desktop and other MCP clients can also be configured to launch the server.

Install Chromium only when browser rendering is needed:

bash
uvx --from free-search-mcp playwright install chromium

Optional settings control proxy use, cache and download locations, browser-agent behavior, local document access, and download retention. Local document reads are disabled unless SEARCH_MCP_DOCUMENT_ROOT points to a sandbox directory. Downloads are enabled by default and are stored in a cache download directory; downloaded files normally expire after 24 hours.

Tools and capabilities

  • search queries multiple engines, applies filters, and returns ranked, deduplicated links.
  • research searches, fetches top results, and produces a source-backed brief in one call.
  • fetch reads a URL, parses documents, or describes binary and media resources.
  • fetch_batch retrieves up to 20 URLs concurrently.
  • read_doc parses PDFs, DOCX, XLSX, PPTX, EPUB, CSV, code, archives, HTML, text, and Markdown, with pagination support.
  • paper_graph follows a paper's references and citing works and reports retraction or correction notices.
  • cache_search runs FTS5 full-text queries over pages already fetched locally.
  • compare returns excerpts from two to five URLs for side-by-side analysis.
  • extract_structured extracts JSON-LD, OpenGraph, Twitter card, microdata, and RDFa metadata.
  • download saves a URL to a temporary local file.
  • engines exposes the available groups, subgroups, and source names.

Markdown is the default output format; tools that support it can return JSON instead. A separate answer tool appears only when an answer backend is configured.

Limitations and notes

Search freshness is best effort because many HTML results have no parseable publication date. Undated results remain in the response, so dates, prices, deadlines, and other changing facts should be verified by fetching the source and refreshing cached data when necessary.

fetch expects an absolute HTTP or HTTPS URL and is not the right tool for a search query. read_doc is intended for documents rather than normal article pages. Local file access is sandboxed and off by default. Browser rendering adds a Chromium dependency, and some engines or sites may require a proxy when they impose CAPTCHA, login, or IP restrictions.

Read the full README View source on GitHub

Reviews

No reviews yet — be the first to share how it worked for you.

Frequently Asked Questions about Free Search

Free Search is an MCP server that connects AI clients to keyless web search and page retrieval. Its main tools include multi-engine search, one-shot research, URL fetching, document reading, cache search, citation graphs, comparisons, metadata extraction, and downloads. It runs locally and returns Markdown or structured JSON.

Install it through the documented Claude Code or Codex plugin flow, or configure it as an MCP server in Claude Desktop or another compatible client. The package uses uv and does not require a source checkout, account, or search API key.

No, its default search sources work without an API key. The server uses a local-first keyless engine pool and can search, fetch, and parse content without a search-service sign-up.

Set SEARCH_MCP_DOCUMENT_ROOT to a directory before starting the server. Local reads remain disabled when the variable is unset, and requested paths must resolve inside the configured root; remote HTTP and HTTPS documents are unaffected.

Yes, both Codex and Claude Code have documented plugin installation commands for Free Search. In Codex, remove an older mcp_servers.search entry if it shadows the plugin server.

Related MCP servers

All in Search & Data Extraction All alternatives
Add a badge to your READMEShow it’s listed — live SVG, Markdown or HTML.

AllMCPs Directory Badge

Full Badge Customizer

Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.

Badge Style:
Live Dynamic SVG PreviewFree Search AllMCPs Directory Badge
Markdown (GitHub README)
[![AllMCPs](https://allmcps.com/api/badge/free-search?style=directory)](https://allmcps.com/mcp/free-search)
HTML Embed
<a href="https://allmcps.com/mcp/free-search"><img src="https://allmcps.com/api/badge/free-search?style=directory" alt="Free Search on AllMCPs" /></a>