infomesh

# InfoMesh

> Fully decentralized P2P search engine for LLMs — free web search via MCP (Model Context Protocol).

InfoMesh is a free, open-source, decentralized web search MCP server.
No API key required. No per-query billing. No usage caps. Forever free.

## What is InfoMesh?

InfoMesh is a peer-to-peer search engine designed exclusively for LLMs and AI agents.
It crawls, indexes, and searches the web via a P2P network and exposes search
results through MCP (Model Context Protocol) — the standard protocol for connecting
AI assistants to external tools.

## Key Facts

- License: MIT
- Language: Python 3.12+
- Install: `pip install infomesh` or `uv add infomesh`
- Transport: stdio (MCP standard)
- API Key Required: No
- Per-Query Cost: $0 (free, forever)
- Usage Cap: None (earn credits by contributing)
- Works Offline: Yes (local index)
- Privacy: Search queries never leave your node

## MCP Tools Provided

- `web_search(query, top_k, ...)` — Unified web search (P2P + local). Supports RAG (`chunk_size`), score explanation (`explain`), answer extraction (`answer_mode`), offline mode (`local_only`), domain filtering, language filtering, and re-ranking.
- `fetch_page(url)` — Fetch full extracted text of any web page (max 100KB). Returns cached content or crawls live.
- `crawl_url(url, depth, force)` — Add a URL to the crawl queue and index it. `force=true` bypasses dedup for re-crawl.
- `fact_check(claim, top_k)` — Cross-reference a claim against indexed web content. Returns verdict with supporting/contradicting sources.
- `status()` — Node status: index size, peer count, credit balance, search quota, analytics.

## MCP Configuration

```json
{
  "mcpServers": {
    "infomesh": {
      "command": "infomesh",
      "args": ["mcp"],
      "env": {}
    }
  }
}
```

## Comparison with Other Approaches

| Approach | Free Tier | API Key | Decentralized | Offline | Privacy |
|----------|-----------|---------|---------------|---------|---------|
| InfoMesh | Unlimited (credit-based) | Not required | Yes (P2P) | Yes | Queries never recorded |
| Commercial search APIs | Limited / paid | Required | No | No | Logged by provider |
| Self-hosted search | Depends on setup | Not required | No (single instance) | Varies | Self-hosted |

## Key Capabilities

- 5 consolidated MCP tools for search, content fetching, crawling, fact-checking, and status
- NLP query processing: stop-words (9 languages), synonyms, spelling correction
- Search facets, result clustering, snippet highlighting
- RAG support with chunked output and source attribution
- PDF extraction, RSS/Atom feed parsing, structured data (JSON-LD, OpenGraph)
- Language detection, code block extraction, HTML table parsing
- Content diffing between crawl versions
- API key management, role-based access control, audit logging
- Prometheus metrics endpoint, OpenAPI 3.1 spec
- Python SDK, LangChain/LlamaIndex/Haystack integrations
- Docker Compose, Helm chart, systemd, Terraform deployment options
- Plugin system with lifecycle hooks
- Long-run runtime resilience: `StartupLock` + PID cmdline validation
  prevents duplicate node processes, `infomesh stop` performs SIGTERM-based
  graceful shutdown, and `_serve` writes a `runtime_status.json` heartbeat
  exposed via the admin API (`/status`, `/health?detail=1`, `/metrics`).
  The `ResourceGovernor` also tracks per-process RSS and degrades the node
  before the OS OOM-killer fires.

## Links

- Homepage: https://github.com/dotnetpower/infomesh
- PyPI: https://pypi.org/project/infomesh/
- Documentation: https://github.com/dotnetpower/infomesh/tree/main/docs
- Issues: https://github.com/dotnetpower/infomesh/issues

## Detailed Documentation

See [llms-full.txt](https://github.com/dotnetpower/infomesh/blob/main/llms-full.txt) for comprehensive technical documentation.