# Frontier Deal Screener — MCP Server

A **free, self-contained** MCP server that serves a curated quantum / deep-tech
startup dataset (~34 companies) and a weighted diligence rubric, from a neutral
investor/acquirer vantage. **No external API, no API keys, no accounts** — all
data and scoring run locally.

It's the queryable version of the Frontier Deal Screener web tool: instead of a
web page, Claude (or any MCP client) can call it directly — "list the interconnect
startups", "score Classiq", "rate a deal with these numbers".

> Company profiles are **illustrative estimates from public information**, not
> assessments of those companies. Weights, tiers, and posture rules are transparent
> defaults — adjust them to your own thesis (edit `screener_data.py`).
>
Webapp here: https://khorovitz.github.io/quantumdeals/

## Tools

| Tool | What it does |
|---|---|
| `screener_list_companies` | List/filter the dataset by layer, stage, posture, or min conviction |
| `screener_get_company` | Full profile + computed rating + source for one company |
| `screener_rate_deal` | Score arbitrary inputs → conviction, tier, posture, rationale |
| `screener_methodology` | The dimensions, weights, 0/3/5 anchors, tiers, risk model, sources |

## Files

- `quantum_deal_screener_mcp.py` — the MCP server (tools)
- `screener_data.py` — the dataset + scoring logic (dependency-free; edit this to tune)
- `requirements.txt` — the one dependency (`mcp`)

## Setup

```bash
git clone https://github.com/<your-username>/quantum-deal-screener-mcp.git
cd quantum-deal-screener-mcp
python3 -m venv .venv && source .venv/bin/activate     # optional but recommended
pip install -r requirements.txt
python3 quantum_deal_screener_mcp.py                   # runs on stdio; Ctrl-C to stop
```

Requires Python 3.10+.

(There's nothing to see when you run it directly — it waits for an MCP client to
connect over stdio. Point a client at it as below.)

## Connect it to Claude

This is a **local stdio** server. Add it to whichever client you use, replacing the
paths with the absolute paths on your machine (and the venv's python if you made one).

**Claude Desktop / Claude Code** — add to your MCP config (e.g. `claude_desktop_config.json`,
or via Claude Code's MCP settings):

```json
{
  "mcpServers": {
    "quantum-deal-screener": {
      "command": "/absolute/path/to/qds_mcp/.venv/bin/python3",
      "args": ["/absolute/path/to/qds_mcp/quantum_deal_screener_mcp.py"]
    }
  }
}
```

If you didn't make a venv, use your system `python3` as the `command` (make sure
`pip install mcp` ran for that interpreter).

**Test without a client** with the MCP Inspector:

```bash
npx @modelcontextprotocol/inspector python3 quantum_deal_screener_mcp.py
```

## Tune it

Everything lives in `screener_data.py`:
- `COMPANIES` — add/remove companies or change any profile
- `DIMS` — dimension weights and 0/3/5 anchors
- `TIERS` — the tier thresholds
- `decide()` — the posture logic (acquire/partner/invest/watch/pass)

Change those and the tools reflect it immediately on restart — no other edits needed.

## License

MIT — see [LICENSE](LICENSE). The bundled company profiles are illustrative
estimates from public information for research/education, not investment advice
or assessments of the named companies.