Are AI agents
recommending you?
Enter your site. See how often agents show you, who they pick instead, and what to fix. Free.
Paste any of these into your agent — Claude, ChatGPT, Cursor, anything.
Drop any API URL. Watch Prowl read, plan, probe, and score it.
Run a 5-phase LLM benchmark in ~12 seconds.
Analyze · Plan · Onboard · Execute · Interpret. We read the spec, generate tests, run real calls, and normalize to a 0–100 score. Free, rate-limited 5/hr.
Paid calls settle in $PROWL. No invoices. No sales call.
Finding a service on Prowl is free. The things that burn LLM time — a re-benchmark, a migration guide, a workflow plan — cost cents, and they're paid in $PROWL over x402: the endpoint answers 402 with a price, your agent pays on Solana and retries with the signature. That's the whole checkout. Nobody books a demo.
- 01 · PER CALLPay as you goSend
X-Payment-Proof: sol:<tx>with the request. One transfer, one answer. - 02 · CREDITSTop up onceSend $PROWL to
/v1/credits/topup— or pay by card ($5–$500) in the dashboard — and spend it down call by call. No block wait per $0.10. - 03 · PROVIDERSGet paid to benchRun
prowl-bench, deliver benchmarks, keep 70% of what vendors pay for them. Withdraw to your wallet.
DRg2EnkqTNFVnBegv1KReGTWs1cGBNCfyyUnY6bkpump$PROWL is how you pay for Prowl, not an investment product. We make no claims about its price, and neither should anyone quoting us.
HTTP/1.1 402 Payment Required { "x402Version": 1, "accepts": [{ "network": "solana-mainnet", "asset": "DRg2En…pump", "extra": { "proofHeader": "X-Payment-Proof" } }] }
Scores come from reality. Not decks.
Ground-truth benchmarks
Weather against NOAA. Geocoding against OSM. Translation against BLEU. You can't game the sources.
Probed from 3 regions
p50, p95, p99 — every five minutes. Regional latency maps included. No handwaving about "fast".
Drift detection
We diff your OpenAPI on every crawl. Added, removed, renamed — breaking changes flagged before agents break.
Self-onboarding
Can an agent register without a human? Onboarding strategies tested: email, crypto wallet, OAuth, public. No CAPTCHAs, no KYC.
The top of the prowl, right now.
Your API has an ASO score whether you claim it or not.
Unclaimed services cap at 75/100. Verify a domain (DNS TXT, .well-known, or meta tag) to unlock the full range and earn a Verified badge.
Every API an agent can find, on one map.
Two sides of the same network.
One endpoint. The entire SaaS ecosystem.
Instead of learning 500 APIs, learn one. Register, discover, consume. We do the reading so your context doesn't have to.
- Register. POST /v1/auth/agents/register — returns an ak_ key.
- Discover. GET /v1/discover?q=payments&sort=score
- Trust. GET /v1/metrics/{id}/verified — benchmarks vs ground truth.
- Report. POST /v1/feedback — close the loop, earn a discount.
# one-liner. free. no key needed for GET /discover curl prowl.world/v1/discover?q="geocoding"&sort=score \ -H "X-Agent-Key: ak_..." → 12 results, ranked · p95 latency · verified badge
Agents can't find you. Your ASO score says why.
SEO made sites findable by humans. ASO makes APIs findable by agents. Free to list, pay to prove you're good.
- Get crawled. If your API is public, you're already indexed.
- Claim. Prove ownership via DNS, well-known file, or meta tag.
- Benchmark. First one free. Re-runs $1.00 — pay in $PROWL.
- Fix. Audit report points at exactly which dimension is dragging your score.
# audit report, free for vendors GET prowl.world/v1/audit/{service_id} → breakdown.schema_stability: 4.2/10 → fix: maintain backward-compat aliases ≥ 90d → impact: +8 points
Design endpoints agents can actually use.
Score an OpenAPI fragment before you ship it.
Parseability, token bloat, error clarity, auth simplicity, schema gotchas — graded against the same priors Prowl uses to benchmark live APIs. Concrete rewrites included. 3 reviews/day free without an account; 10/day logged-in.
# score a spec, free tier or $0.02 via x402 curl prowl.world/v1/endpoint/review -X POST \ -d '{"spec_url":"https://api.you/openapi.json"}' → parseability 6.2/10 · token-bloat 4.1/10 → fix: paginate with `cursor`, not `offset`+`limit`
See exactly which model breaks on your endpoint.
Claude and GPT-4o hit the URL with the spec you ship. We return per-model traces — "Claude parsed first try, GPT-4o retried because pagination uses next_cursor."
# live test against a real URL curl prowl.world/v1/endpoint/test -X POST \ -d '{"spec_url":"...","target_url":"...","method":"GET"}' → claude: 1 call, 200 OK → gpt-4o: 3 retries, schema confusion on `cursor`
No dashboard. Just JSON.
We optimize for the next token, not the next tab.
Prowl is an API. The only HTML on prowl.world is this page — and it is designed to get out of the way. Claude, GPT, Mistral, or a custom orchestrator parses the response in milliseconds.
Tabs swap the endpoint. Response is the whole UI.
The agent economy runs on discovery.
Be the thing it finds.
— services indexed. — benchmarks in the last 24 hours. Free for agents. One curl from a key.