# jev-legal-clause-mcp

> 🇨🇳 [中文文档](README.zh-CN.md)

> Let your agent flag contract-clause risks with **calibrated-probability structured decisions** instead of parsing free-form LLM text.
> Powered by [TypeSafe Jev](https://typesafe.ai) (System One decision model).

A **single-purpose MCP tool** that turns a contract excerpt into calibrated risk annotations — each of 10 default risk classes returns a
**presence probability (0..1)**, plus an overall risk level and a "suggest human review" flag. Your agent gets
**verifiable risk labels**, not a legal opinion.

```
npm install && npm run build
node dist/index.js doctor      # print dependency/mode diagnostics
```

Connect `node dist/index.js` in your MCP client and use the `flag_clauses` tool.

---

## ⚠️ This is not legal advice

This tool is a **contract first-pass / triage aid**. It uses a decision model to attach probability labels to clauses and cannot replace a lawyer.
**Always have a qualified human review any contract before signing.** A high probability on one risk class **does not mean** the rest of the clauses are safe.

---

## Tool: `flag_clauses`

**Input**
| Param | Type | Description |
|---|---|---|
| `text` | string | The contract excerpt to scan (a clause / a section / the whole thing) |
| `types` | string[]? | Optional, custom risk classes; defaults to the built-in 10 |

**Default 10 risk classes**
| Class | Meaning |
|---|---|
| `indemnification` | Indemnification / hold-harmless obligations |
| `non_compete` | Non-compete / non-solicit |
| `confidentiality` | Confidentiality obligations |
| `termination_penalty` | Early-termination penalty |
| `auto_renewal` | Auto-renewal (renews unless notice given) |
| `liability_cap_absent` | Missing liability cap (unlimited liability) |
| `ip_assignment` | IP ownership / assignment |
| `governing_law` | Governing law / dispute venue |
| `payment_terms` | Payment-term risk (terms / penalty interest) |
| `personal_guarantee` | Personal guarantee (joint & several liability) |

**Output fields**
| Field | Meaning |
|---|---|
| `clauses` | Per class `{ category, label, present_prob }` |
| `flagged` | Classes with `present_prob > 0.7`, sorted by probability descending |
| `highest_risk` | The highest-probability item in `flagged` (null if none) |
| `overall_risk` | `{ score, label, confidence }`, levels low / moderate / high / severe |
| `review_recommended` | Whether any flagged item exists or overall risk ≥ high |
| `mocked` | Whether running in mock mode |

**Gate logic**
- Any class with `present_prob > 0.7` → enters `flagged`
- `flagged` non-empty **or** `overall_risk.score ≥ 2` → `review_recommended = true`

Before any action, check `review_recommended` + `overall_risk.confidence` first.

---

## Example

```json
{
  "clauses": [
    { "category": "indemnification", "label": "赔偿 / 补偿义务（hold harmless）", "present_prob": 0.92 },
    { "category": "governing_law", "label": "管辖法律 / 争议解决地", "present_prob": 0.81 },
    { "category": "non_compete", "label": "竞业限制", "present_prob": 0.12 },
    "…"
  ],
  "flagged": [
    { "category": "indemnification", "label": "赔偿 / 补偿义务（hold harmless）", "present_prob": 0.92 },
    { "category": "governing_law", "label": "管辖法律 / 争议解决地", "present_prob": 0.81 }
  ],
  "highest_risk": { "category": "indemnification", "label": "赔偿 / 补偿义务（hold harmless）", "present_prob": 0.92 },
  "overall_risk": { "score": 3, "label": "severe", "confidence": 0.83 },
  "review_recommended": true,
  "mocked": false
}
```

---

## Zero-key / local-first

- No `TYPESAFE_API_KEY` set → automatically enters **mock mode** (deterministic offline stub, zero-config for CI / demos)
- `JEV_MCP_MOCK=1` → force mock
- Set `TYPESAFE_API_KEY` → call the real Jev API (default `jev-latest` @ `https://api.typesafe.ai/v1/systemone`)

**Environment variables**
| Variable | Default | Description |
|---|---|---|
| `TYPESAFE_API_KEY` | — | Required for real calls; blank falls back to mock |
| `JEV_MODEL` | `jev-latest` | Model name |
| `JEV_BASE_URL` | Official endpoint | Override for self-hosted / proxy |
| `JEV_MCP_MOCK` | `0` | `1` forces mock |
| `JEV_MCP_TIMEOUT_MS` | `30000` | Per-request timeout |

---

## Extremely low cost

A single decision is just one Jev call (10 noul + 1 score), state truncated at 60k characters,
tens of thousands of tokens per call, with no LLM calls at all. Think of it as "one cheap gate in your contract pipeline".

---

## Tests

```bash
npm test     # smoke + MCP protocol handshake (mock mode, no key needed)
```

---

## License

MIT