Every time your agent signs up for a service, it agrees to terms no one reviewed — arbitration waivers, data-training licenses, unilateral price changes. Fineprint reads them first: a 0–100 risk score, a pass / flag / block verdict, every clause quote-grounded in the vendor's own text. Gate the vendor before the agent commits.
One email, free, 25 scans/month. The key shows once — get it here.
POST /v1/scans with any terms URL. Verdict back in about a minute.
Set a POST /v1/monitors on vendors you depend on. We alert you when terms change.
One job: turn legal prose into auditable risk data. The model extracts. It never judges.
Pulls the document, strips navigation and chrome, hashes content. Unchanged documents are never re-scanned — and unchanged monitors never bill.
Chunked extraction against a 21-category taxonomy. Every clause must be quote-grounded in the source, or it is dropped.
Risk math and pass/flag/block verdicts execute with zero model involvement. Same document, same score, every run.
An agent's job: onboard a vendor — create the account, accept the terms, start using it. Its standing policy: never automate against terms that prohibit it; official channels only where restricted; when permission is unclear, stop and ask the human. Pick a vendor and run it. The agent is simulated — every score, verdict, and quote below is live data from the public leaderboard.
Reproduce it in your own terminal: python3 demo/agent_onboarding_demo.py from the public fineprint-skill repo — same policy, same live data. In production this check is one MCP tool call (fineprint_check_agent_access) or one API request, made before your agent acts. Fineprint returns the decision and the evidence; your application enforces it.
Every major vendor, scored on the same taxonomy. Compare terms before your agent accepts them — or before spending a scan. Free to read, no key required.
A high score isn't always a reason to walk away — sometimes it's a reason to negotiate harder and watch closer.
Every vendor also gets an agent-friendliness verdict: does their ToS even allow your agent to use the service? Plenty of terms ban bots and "automated access" outright — and the violation lands on you, not your agent.
| Service | Score | Level | Agent | Top flags | Watch |
|---|---|---|---|---|---|
| Loading live scores… | |||||
Hit Watch on any row, leave an email, and we monitor that vendor's terms for you: unchanged checks are free, and material changes land on your personal watchlist page.
Risk score: 0 = cleanest terms, 100 = most rights reserved against you.
Each vendor's public terms are extracted against a fixed 21-category taxonomy — arbitration, data training, termination, bot bans, agent delegation, and the like. Every finding must be quote-grounded in the source; no quote, no clause. Scoring is deterministic severity-weighted math critical ×25 · high ×12 · medium ×5 · low ×2 — no language model touches the score, so the same document always scores the same. 0–24 low · 25–49 medium · 50–74 high · 75–100 critical
Pick a vendor. Full report, worst clauses first, every claim anchored to a manually verified quote. Served from the public leaderboard — no key, no quota.
// LIVE VENDOR REPORTS POPULATE AT LAUNCH — THIS DEMO READS THEM IN REAL TIME
All reports (no JS needed): GitHub · Zoom · Stripe · X · OpenAI · Anthropic · Dropbox · Google · Basecamp · DuckDuckGo · Signal · Figma
Terms change — usually quietly, sometimes against you. Set a monitor on any vendor's terms and we re-check on your schedule.
Content-hash comparison first. If nothing moved, the check costs nothing — no quota, no LLM call.
New text triggers a full re-scan: clause-level diff, score delta, and what got worse.
Webhook when it matters — score up 10+, a new high/critical clause, or a worsened verdict.
Built for machines, not browsers. Two calls to first verdict. Full reference: interactive API docs · ReDoc · OpenAPI JSON
Twelve tools — scan URL, scan text, batch scan, compare scans, report, leaderboard, quota, watch terms, list monitors, custom policies, negotiation playbooks. Drops into Claude, Cursor, ChatGPT.
POST /v1/customers issues a key. POST /v1/scans returns a verdict. That is the integration.
/llms.txt and /.well-known/fineprint.json describe the full service surface for agents.
If it is not in the taxonomy, it does not ship. Off-taxonomy output is dropped, not passed through.
binding_arbitrationForced arbitration, no courtclass_action_waiverNo collective lawsuitsunilateral_modificationTerms changeable at willunilateral_price_changePrices rise without noticeauto_renewalRenews unless cancelleddata_collection_scopeWhat is collected on youdata_sharing_resaleYour data, sold onwarddata_training_useYour data trains modelsdata_retention_deletionCan you delete? Ever?liability_capHow little they owe youindemnificationYou cover their legal coststermination_at_willCut off without causeaccount_suspensionSuspended without noticeip_assignmentWho owns what you makejurisdiction_venueWhere you would have to suesla_absenceNo uptime commitmentusage_capsThrottling and limitsbot_automation_banBots banned outrightaccount_delegationCan an agent act for you?ai_agent_termsExplicit AI-agent rulesautomation_permittedAutomation allowed—Closed set. Additions versioned.Monitoring is the retention layer: free tier watches 3 vendors, Pro watches 50, Team watches 200, Business pilot watches 500. Unchanged checks never consume quota.
No key? Agents will be able to pay per scan: x402 pay-per-scan in USDC ($0.25/scan) — rolling out; contact us to enable. POST /v1/scans/x402
One email, one key. 25 scans/month free, no card. The key is shown exactly once — store it somewhere safe.
Agents: no email? POST /v1/customers/anonymous mints a 5-scan trial key — or x402 pay-per-scan at POST /v1/scans/x402 (USDC, rolling out soon), no key at all.
Prefer the terminal? curl -H "X-API-Key: $KEY" /v1/me
// KEYS ARE FREE — ABUSE IS RATE-LIMITED BY THE SAME QUOTAS THAT GOVERN THE API
POST /v1/monitors) and receive a webhook alert when the change is material: score up 10+, new high/critical clause, or worsened verdict. Stale leaderboard entries are flagged after 30 days.GET /v1/scans/{id}/report.pdf renders a branded audit PDF — verdict, score, every finding with its quote-grounded evidence, methodology, and evidence provenance (content hash, scan id, timestamp). Multi-vendor packs via POST /v1/reports/audit-pack.pdf add a ranking and a category-by-category comparison matrix. Download a sample report.GET /v1/leaderboard needs no key and consumes no quota. The board is curated by the Fineprint team — your scans stay private by default.