by Script Master Labs LLC
Fact-check a claim against live web sources, with quoted evidence
A useful gate for agents that must not repeat unchecked claims, with a free preview and receipts you can reopen. Treat its verdicts as unbenchmarked, because the vendor says so, and confirm the price first, since its pages quote two different amounts for one check.
Verdicts that can be reopened later, because each paid check is sealed as a receipt whose quotes were machine-checked against the source text.
Main trade-off
An unsettled price and no measured accuracy.
One paid tool charges per check in USDC or another accepted asset, with no subscription, account or key. The other four tools are free. No licence or repository was found for the service.
Exposed as MCP tools over streamable HTTP; the agent calls fact_check_claim by name and answers the x402 payment challenge, or calls the free tools with no payment.
The permissions this connector asks for, as its own documentation states them.
5 actions an agent can call. Open one for its parameters and an example.
Example call
{"name":"fact_check_claim","arguments":{"claim":"The Eiffel Tower is located in Paris.","sources":["https://en.wikipedia.org/wiki/Eiffel_Tower"]}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| claim | string | Required | The claim to check, as one statement or a short paragraph. |
| sources | array | Optional | Public HTTPS URLs to check against, at most 8; the service searches the web when fewer than 2 are usable. |
| sub_claims | array | Optional | Your own split of the claim into at most 4 sub-claims; otherwise the text is split into sentences. |
Returns
| Field | Type | Description |
|---|---|---|
| verdict | string | SUPPORTED, CONTRADICTED, MIXED or UNVERIFIABLE. |
| coherence | number | The agreement score behind the verdict. |
| receipt_id | string | The id of the sealed evidence receipt. |
| settlement | object | The settlement transaction for the payment. |
Example call
{"name":"fact_check_preview","arguments":{"sources":["https://www.britannica.com/place/Mount-Everest"],"claim":"Mount Everest is the highest mountain above sea level."}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| sources | array | Optional | Public HTTPS URLs to fetch instead of searching, at most 8. |
| claim | string | Required | The statement whose sources you want to see before paying. |
Example call
{"name":"fact_check_status","arguments":{}}Returns
| Field | Type | Description |
|---|---|---|
| status | string | Service state, such as ok. |
| jury_configured | boolean | Whether the juror model is configured. |
| receipts_count | number | How many receipts are stored. |
| networks | array | The payment networks accepted for the paid check. |
Example call
{"name":"fact_check_methodology","arguments":{}}Example call
{"name":"fact_check_receipt","arguments":{"receipt_id":"cg_mvuzfnux_b4319b13"}}Parameters
| Parameter | Type | Needed | Description |
|---|---|---|---|
| receipt_id | string | Required | The receipt id returned by a paid check. |
Returns
| Field | Type | Description |
|---|---|---|
| verdict | string | The verdict stored with the receipt. |
| sources | array | The sources and per-family verdicts with validated quotes. |
An MCP fact-checking service from Script Master Labs that splits a claim into sub-claims, fetches live web sources, has AI jurors judge each source family with verified quotes and returns a coherence-gated verdict. It offers five tools, one paid through x402, and seals each paid check as a receipt.
Quote-verified evidence
Every juror judgment must carry a verbatim quote that the service checks against the fetched source text before it counts.
Coherence-gated verdicts
A verdict needs an agreement score of at least 0.80 across at least 2 independent source families, and anything less is MIXED or UNVERIFIABLE.
Free preview
A no-cost call shows the sub-claims, the sources and their families before any payment.
Sealed, re-fetchable receipts
Each paid check is stored as a receipt that anyone with its id can fetch for free.
Five payment rails
The paid check accepts USDC on Base, Arc or Solana, XRP on XRPL and XAH on Xahau.
Honest UNVERIFIABLE
A claim that no source addresses is returned as UNVERIFIABLE instead of being guessed.
Verify a claim before an agent repeats it
An agent sends a statement to the paid check and republishes it only if the verdict is SUPPORTED with quoted evidence.
Preview what would be checked, for free
A cautious agent calls the preview first to see the sub-claims, the sources and the source families before it pays.
Audit a verdict later from a receipt
A reviewer fetches the sealed evidence pack by receipt id to read the per-family verdicts and verified quotes.
Gate publishing on evidence
A content pipeline blocks any sentence whose check comes back CONTRADICTED, MIXED or UNVERIFIABLE until a human looks.
The nearest AI agent skills on Hokai for the same job.
| Skill | Works with | Connects by | Checked |
|---|---|---|---|
| -- | MCP | 17 Nov | |
| Claude · Cursor | MCP | 26 Oct | |
| Claude · Cursor | REST API | 11 Oct | |
| Claude · ChatGPT · Gemini | REST API | 11 Oct |
The vendor's product page, dated 30 October 2026, says one check costs 0.25 USDC on Base with no subscription, account or key. The live tool description and its own status reply, read on 17 November 2026, quote 0.03 USDC for that check, payable in USDC on Base, Arc or Solana or in XRP or XAH at the live spot rate. Treat the 402 challenge your agent receives as the price that counts.
Four of its five tools are free: the preview, the status call, the methodology and the receipt fetch. Only the fact-check call is paid. No licence, repository or package was found for the service, so it should be treated as a closed hosted service.
Clients that implement the streamable HTTP transport of MCP can connect to its endpoint, which answered initialize and tools/list calls when tried on 17 November 2026. The vendor also offers a plain HTTP mirror for clients without MCP, and names no specific assistants.
SUPPORTED or CONTRADICTED requires agreement of 0.80 or higher, all pointing one way, from at least 2 independent families. MIXED covers scores from 0.50 to 0.79, and UNVERIFIABLE is anything else, including a claim that fewer than 2 families address. The vendor says it never guesses.
An unpaid request receives only the x402 payment challenge. The vendor states that failed calls settle nothing, and the tool description says a call fails before payment is requested when the jury is not configured, so nobody pays for a check that cannot run.
ClaimGate is a fact-checking service for AI agents from Script Master Labs LLC, the maker of OMNIGATE and ScriptMaster SEO Command. It is a skill because an agent calls it as an MCP tool before relying on a statement: it adds a way to test a claim against live web sources and get back a verdict with sealed evidence. The vendor's product page and price list are dated 30 October 2026, and the live status call reported version 1.0.0 when called on 17 November 2026.
An agent sends a claim, optionally with up to eight public source URLs. The service splits the claim into sub-claims, fetches the sources, groups them into independent source families and has AI jurors it calls Jev judge each family, with every judgment tied to a verbatim quote that is checked against the fetched text. A coherence gate then returns SUPPORTED, CONTRADICTED, MIXED or UNVERIFIABLE and stores a receipt. Calls go to an MCP endpoint over streamable HTTP, or to a plain HTTP mirror, and a paid call first returns an x402 payment challenge.
Any MCP client that supports streamable HTTP can call it. It fits an agent that must verify a claim before republishing it, a research agent that wants quoted evidence per source family, and a reviewer who wants to reopen the evidence later from a receipt id. It is filed under search skills.
Do not use it for claims that no public web page addresses, which come back UNVERIFIABLE by design, or when you need a measured accuracy figure: the vendor labels its verdicts calibrated false, which means they were not tested against a labelled set of fact-checks. No other fact-checking skill on Hokai does the same job, so there is no listed alternative to weigh it against.
On 30 October 2026 the vendor ran three proving checks from its own wallet, a supported claim, a false claim that came back UNVERIFIABLE and an invented one, and labelled them self-proof with receipt ids that anyone can fetch.
Protocol, authentication, permissions and the call contracts on this page are read from Script Master Labs LLC's own documentation. The verdict is Hokai's reading of them, not the maintainer's claim.