<img src="assets/nerq-logo-400.png" alt="Nerq" width="120" align="right" />

# Nerq MCP Server

**Official first-party MCP server for the Nerq catalog** ([nerq.ai](https://nerq.ai)).

Nerq indexes 6.8M+ software entities, AI tools, packages, and **41,368 MCP servers**, each with an
independent Trust Score. This Model Context Protocol server exposes Nerq's trust judgments as agent
tools, so an agent can check trust *before* it acts.

- **Hosted endpoint (Streamable HTTP):** `https://mcp.nerq.ai/mcp`
- **Server card:** [`https://mcp.nerq.ai/.well-known/mcp/server-card.json`](https://mcp.nerq.ai/.well-known/mcp/server-card.json)
- **Registry namespace:** `ai.nerq/mcp` (DNS-verified on `nerq.ai`)
- Free, no authentication.

This is Nerq's own server for its own catalog — first-party, not a third-party wrapper. The catalog
and the trust judgments are **hosted**; this repository is a thin client bridge and a manifest. It
holds **no data, no database access, and no business logic**.


## Install

[![Add to Cursor](https://img.shields.io/badge/Add%20to-Cursor-0f172a?logo=cursor&logoColor=white)](cursor://anysphere.cursor-deeplink/mcp/install?name=nerq&config=eyJ1cmwiOiJodHRwczovL21jcC5uZXJxLmFpL21jcCJ9)

Or point any Streamable-HTTP MCP client at `https://mcp.nerq.ai/mcp` (see **Using it** below).

## Tools

| Tool | Signature | Returns |
|---|---|---|
| `is_safe` | `is_safe(entity)` | Trust Score, grade, verdict (safe / caution / elevated risk / **not yet scored**), `source_url`. Abstains if the entity can't be confidently resolved. |
| `compare` | `compare(a, b)` | Head-to-head by Nerq Trust Score, with a winner — or `status: "cannot_compare"` if either side can't be confidently resolved. **Nerq abstains rather than compare the wrong entity.** |
| `search_assets` | `search_assets(query, type?, limit?)` | Entities matching `query`, ranked by Trust Score. Optional `type` filter (`npm`, `pypi`, `crates`, `mcp_server`, …). |
| `find_mcp_servers` | `find_mcp_servers(capability?, limit?)` | MCP servers matching a capability keyword, ranked by Trust Score, from Nerq's index of 41,368. |

### Design guarantees

- **Never fabricates.** Resolution is exact — no fuzzy matching. On a miss the tools abstain
  (`found: false` / `cannot_compare`) rather than return a wrong entity: an agent acting on a
  fabricated match is worse than no answer.
- **Never a numeric zero for the unscored.** An entity with no score returns `"not yet scored"`,
  never `0/100`.
- **Provenance on every response.** Every result carries a `source_url` back to the Nerq page it came
  from, so the judgment is citable.

## Using it

**Recommended — connect directly to the hosted server** (Streamable HTTP), no install:

```
https://mcp.nerq.ai/mcp
```

```python
import asyncio
from mcp.client.streamable_http import streamablehttp_client
from mcp.client.session import ClientSession

async def main():
    async with streamablehttp_client("https://mcp.nerq.ai/mcp") as (r, w, _):
        async with ClientSession(r, w) as s:
            await s.initialize()
            print(await s.call_tool("is_safe", {"entity": "langchain"}))

asyncio.run(main())
```

**stdio clients** — this repo's `server.py` is a thin bridge that forwards stdio to the hosted
endpoint (no data, no secrets):

```bash
pip install -r requirements.txt
python server.py            # bridges stdio -> https://mcp.nerq.ai/mcp
```

## About Nerq

Nerq is an independent, quantitative trust layer for software and the machine economy. Trust Scores
are computed from independently measured dimensions (security, maintenance, popularity, compliance,
…) and are machine-readable. Learn more at [nerq.ai](https://nerq.ai).

## License

MIT — see [LICENSE](LICENSE).