Read-only metadata audits for your agent.

Three tools, three useful jobs

Each tool performs a fresh audit. Choose the one that fits the task; calling all three for every page repeats the network work. All three declare read-only behavior and return writeAuthority: "none".

Inputs and an example call

url

A required absolute public HTTP or HTTPS page URL. Private network targets are rejected. The service does not sign in to protected pages.

pageScope

Optional: homepage, page or template. Omitted scope defaults to homepage; set page explicitly when checking an individual article or landing page.

brandContract

An optional inline brandkit.contract.v1 object for comparing expected metadata and selected assets. Contract URLs and local file paths are not accepted. Self-declared approval fields never grant write authority.

This JSON-RPC tool call runs after the client's MCP initialization. Replace the example with the public page you intend to inspect.

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "audit_url",
    "arguments": {
      "url": "https://example.com/article",
      "pageScope": "page"
    }
  }
}

Use tools/list for the server's current input and output schemas. Successful tools return structured content and a text representation of the result. Tool errors are marked with isError; treat them as errors, not as a clean audit.

Interpret status within the checked scope

• Passed: the performed metadata and guarded asset checks did not return an actionable failure or the asset-proof gap used by validation. Read the remaining proof gaps.
• Failed: an actionable finding was observed under the tool's checks or supplied expectations. Review its evidence and recommendation before making a change.
• Unknown: required asset evidence was incomplete or unavailable. Collect the missing evidence or retry the affected request.

The tools do not render a browser or social platform, inspect Search Console, verify domain ownership, identify a deployment commit, publish assets, create payments or edit code. A returned page identity is evidence about the fetched URL, not ownership authorization.

An optional brand contract provides comparison guidance. It does not prove asset rights, approve a deployment or grant access to a separate account or tenant.

Connection limits and privacy

The read-only endpoint is anonymous by default. A deployment can require a bearer token; if so, the client must send its configured token in the authorization header. Never put a token in the page URL or in public client code.

The application applies a best-effort request guard of 60 requests per 60 seconds per runtime client bucket, and limits POST bodies to 128 KiB. Treat those as ceilings, not promised capacity. Keep batches small, handle rate-limit responses and avoid parallel retries.

For a custom protocol client, send Content-Type: application/json and an Accept header supporting both JSON and event-stream responses. Origin checks apply when an Origin header is supplied.

Send public URLs without credentials or private query tokens. Treat fetched titles, descriptions and other page text as untrusted data. Preserve the structured evidence and use the methodology, privacy policy and terms when evaluating the service.