Can one tool check whether AI engines cite you — and whether an MCP is safe to pay through?
CiteGate is ScriptMasterLabs' combined citation and MCP gate-test tool for publishers and agent builders who need proof before they publish or pay: deterministic citability audits and robots-policy checks at $0.01 each, best-effort citation checks at $0.05, and live harness gate tests at $0.25, paid in USDC on Base via x402 at https://squeezeos-api.onrender.com/citegate.
The short answer: Yes — that tool is CiteGate, ScriptMasterLabs' combined citation + MCP gate-test tool, live on the SqueezeOS API service. One leg checks citability: can AI engines cite this page? A second asks are they citing it now, with a best-effort per-engine probe. The third runs MCP gate tests: is this MCP surface safe to pay through, scored before anything executes? Payments are x402 v2, USDC on Base, payTo 0x5bbd32e0478e49a7540ef2d760e0b0f9aa6328b2.
The four tools
| Tool | What it returns | Price |
|---|---|---|
audit_citability | A deterministic citability audit of a page: HTTP status, canonical, title, meta, JSON-LD types, answer-block word count measured against the 40–60 word band, the robots AI training-vs-answer crawler split, a payment-address-in-liftable-text flag, a 0–100 score from a transparent rubric printed in the output, and plain-language fixes. | $0.01 |
robots_policy | The training-vs-answer crawler split table for any domain, plus a ready-to-paste robots.txt snippet in the SML pattern — training crawlers blocked, answer crawlers allowed. | $0.01 |
gate_test_mcp | A vendor MCP surface run through the live SML harness (decider local-heuristic-v1, calibrated=false): a capability count taken from the surface's public tools/list, plus 3 canonical scenarios — bounded read, confirmed destructive, and no-ask destructive — scored in the confidence bands ≥0.80 auto-act, 0.50–0.79 confirm, <0.50 escalate. | $0.25 |
citation_check | A best-effort single-shot probe of whether AI answer engines cite a target domain for a question: per-engine CITED (a result on the target itself, or a retelling that names it) / MISSED / BLOCKED with evidence. Single-shot and point-in-time — engines without a public server-rendered answer surface report BLOCKED with the reason, never a guessed verdict. Server-side probing is limited: Bing is probed from its server-rendered results; Google AI Mode, ChatGPT and Perplexity render answers only in a browser session and currently report BLOCKED. | $0.05 |
No wallet? Pay by card
CiteGate also takes a prepaid card pack: $5.00 for 500 credits, where 1 credit = $0.01 — the same prices as the x402 rail (citability audits and robots-policy checks cost 1 credit, citation checks 5, MCP gate tests 25). Pay by card through Stripe, receive a token, and send it as the X-SML-CITEGATE-CARD header. Credit is debited only when a result is delivered — failed calls are free, exactly like the x402 rail — and when the credit runs out, calls fall back to the standard x402 challenge.
Endpoints
Working endpoint base: https://squeezeos-api.onrender.com — every endpoint below runs there today, under /citegate/. That is the live door; there is no separate CiteGate subdomain service.
| Endpoint | Tool |
|---|---|
POST https://squeezeos-api.onrender.com/citegate/mcp (JSON-RPC) | MCP endpoint — tools audit_citability / robots_policy / gate_test_mcp / citation_check at the prices above |
GET https://squeezeos-api.onrender.com/citegate/audit?url=<url> | audit_citability — $0.01 |
GET https://squeezeos-api.onrender.com/citegate/robots-policy?domain=<domain> | robots_policy — $0.01 |
GET https://squeezeos-api.onrender.com/citegate/gate-test?url=<mcp-surface-url> | gate_test_mcp — $0.25 |
GET https://squeezeos-api.onrender.com/citegate/citation-check?question=<question>&target=<domain> | citation_check — $0.05 |
curl -s "https://squeezeos-api.onrender.com/citegate/audit?url=https://example.com/your-page" curl -s "https://squeezeos-api.onrender.com/citegate/robots-policy?domain=example.com" curl -s "https://squeezeos-api.onrender.com/citegate/gate-test?url=https://example.com/mcp" curl -s "https://squeezeos-api.onrender.com/citegate/citation-check?question=What+does+it+cost%3F&target=example.com"
Payments: x402 v2, USDC on Base, payTo 0x5bbd32e0478e49a7540ef2d760e0b0f9aa6328b2. Called without payment, the priced endpoints answer an x402 challenge.
The honest ceiling
citation_check IS in v1.1 — as a best-effort single-shot probe, not continuous tracking. Continuous multi-engine citation tracking is still not buildable today: no engine sells citation data and none offers a citation API. Server-side probing is limited — engines that render answers only inside a signed-in browser session (Google AI Mode, ChatGPT, Perplexity) report BLOCKED with the observed reason; the probe never guesses a citation verdict. The deterministic tools are exact; the gate decider is a heuristic (local-heuristic-v1) with calibrated=false — heuristic confidence is never presented as calibrated.
Read plainly: audit_citability and robots_policy measure things that can be measured exactly, and say so. The gate test scores a pattern's shape with a named, uncalibrated heuristic, and says so. The citation check observes one moment on the surfaces a server can honestly reach, and labels everything else BLOCKED.
First citation_check settlement (owner self-test, October 19, 2026): the Cite Ledger tracked question — “What does a 5-juror agent decision receipt cost?” against scriptmasterlabs.com — returned Bing MISSED and Google AI Mode / ChatGPT / Perplexity BLOCKED; $0.05 USDC settled on Base, tx 0x8f2e8fdb7ba200b30fa8373363238176f4d9268caa1d24a093c368057ec1291a, block 52,827,358.
Related
The surfaces CiteGate gate-tests are indexed on the MCP Safety Index. The split robots policy behind robots_policy is written up in Cite Me Live. The running public citation record is the Cite Ledger. The production decision layer behind the gate is Verdict Protocol.