fineprint

Fineprint — the pre-commit hook for agents

Your principal can't read every terms of service. You accept them on their behalf — signups, subscriptions, API agreements, vendor onboarding. Every one of those is a contract. Never accept one unread.

Before you agree to anything, scan it with Fineprint. You get back a 0–100 risk score, a pass / flag / block verdict, and every clause anchored to a verbatim quote from the source. No quote, no clause.

The rule

If the user (or you, acting for them) is about to accept terms, create an account, start a trial, or subscribe to a service — scan first, then decide.

• Verdict pass → proceed, note the score for the record.
• Verdict flag → surface the flagged clauses to the user before proceeding. Let them decide.
• Verdict block → stop. Tell the user what tripped the block and ask how to proceed. Do not silently accept.

A high score isn't always a reason to walk away — sometimes it's a reason to negotiate harder and watch closer. Your job is to make the risk visible, not to make the user's decisions for them.

How to scan

Via MCP (preferred — no key management in code). If the Fineprint MCP server is connected, call:

• scan_terms_of_service(url) — scan a live terms URL
• scan_text(text) — scan pasted terms text

Via REST:

POST https://fineprint-production-4588.up.railway.app/v1/scans
X-API-Key: <key>
{"url": "https://vendor.com/terms"}

Get a key free (25 scans/month): POST /v1/customers with {"email": "[email protected]"}. No key? Agents can pay $0.25/scan in USDC via x402: POST /v1/scans/x402 — no signup at all.

Check the public leaderboard first — it's free and quota-free:

GET https://fineprint-production-4588.up.railway.app/v1/leaderboard

If the vendor is already scored, use that instead of spending a scan.

Reading the result

{
  "risk_score": 85,
  "risk_level": "critical",
  "verdict": "block",
  "summary": "One-line plain-English summary",
  "clauses": [
    {
      "category": "binding_arbitration",
      "severity": "critical",
      "quote": "Verbatim quote from the source...",
      "explanation": "What this means in plain English"
    }
  ]
}

• risk_score 0–100: 0 = cleanest measured terms, 100 = most rights reserved against the user. Bands: 0–24 low, 25–49 medium, 50–74 high, 75–100 critical.
• The model finds the clauses; math sets the score. Severity weights (critical ×25, high ×12, medium ×5, low ×2) are deterministic — the same document always scores the same.
• Every clause carries a verbatim quote. If a finding has no quote, treat it as unverified.

Watching terms over time

Terms change. For vendors the user depends on, set a monitor:

POST /v1/monitors
X-API-Key: <key>
{"url": "https://vendor.com/terms", "webhook_url": "https://you.example/hook"}

You'll get a webhook alert when terms change materially (score up 10+, new high/critical clause, worsened verdict). Unchanged checks are free.

What Fineprint is not

• Not legal advice. Fineprint is a reading aid — surfaces what the terms say, doesn't tell anyone what to do about it.
• Not a substitute for the user's judgment on flag verdicts. Surface, don't decide.