auditFreeDirect CrawlHeld Back (Roadmap)

Audit Site

audit_site(url: string, maxPages?: number)

Fast technical/content health audit.

Crawl health audit returning overall health score (0-100), clustered technical issues, and per-page issues (broken links, missing meta, heading structure).

Input Parameters

2 arguments

Parameter	Type	Requirement	Default / Constraints	Description
`url`	`string`	required	Valid website URL (e.g. https://example.com)	The site to audit.
`maxPages`	`number`	optional	default: `50`Integer between 1 and 200	Page cap for this crawl.

Return Envelope (OFE / 1.0)

Open Fact Envelope

Every response adheres to the strict ofe/1.0 envelope schema, returning verified data, typed facts, entity references, and follow-up tool suggestions:

`data` Payload

Crawl health audit returning overall health score (0-100), clustered technical issues, and per-page issues (broken links, missing meta, heading structure).

`coverage` & Freshness

Reports returned count vs total items, observation timestamp as_of, and scope notes. Evaluated live per request.

`resources` (Dataset Exports)

If a query yields high row counts (e.g. >1,000 queries in Search Console), full unpaginated tables are persisted to R2 and linked as an mcpseo:// URI for follow-up retrieval via export_dataset.

`next_actions`

Provides suggested follow-up tool calls with pre-filled arguments so your AI agent can navigate from discovery to detailed inspection autonomously.

Emitted Facts & Entities

Knowledge Graph

Typed Facts

Facts emitted in the facts[] array with provenance receipts:

Fact Type	Claim Description	Emitted Data Fields
`audit.site_health`	Aggregated site health score, pages scanned count, and total issues.	`scorepages_scannedtotal_issues`
`audit.issue_cluster`	Clustered technical issue group with severity and affected count.	`cluster_idissue_typecountseverityaffected_urls`
`audit.crawl_issue`	Specific issue observed on a single crawled URL.	`page_urlissue_typeseveritydetails`

Registered Entities

Entities registered in the entities[] array to establish subject relationships:

domain

Audited domain.

page

Crawled pages.

Request & Response Examples

Live MCP Payloads

Exact JSON schemas transmitted over Model Context Protocol (stdio or HTTP SSE transport):

1. Client Tool Invocationaudit_site request.json
{
  "tool": "audit_site",
  "arguments": {
    "url": "https://example.com",
    "maxPages": 25
  }
}

2. Server Envelope Responseaudit_site response.json
{
  "schema_version": "ofe/1.0",
  "domain": "audit",
  "data": {
    "url": "https://example.com",
    "pages_scanned": 25,
    "max_pages": 25,
    "truncated": false
  },
  "facts": [
    {
      "type": "audit.site_health",
      "subject": [
        "urn:domain:example.com"
      ],
      "data": {
        "score": 92,
        "pages_scanned": 25,
        "total_issues": 3
      },
      "provenance": {
        "source_class": "crawl",
        "method": "self_crawl.site_health",
        "confidence": 0.8,
        "observed_at": "2026-09-24T00:00:00Z",
        "cache_hit": false
      }
    }
  ],
  "entities": [
    {
      "id": "urn:domain:example.com",
      "kind": "domain",
      "label": "example.com"
    }
  ],
  "coverage": {
    "returned": 1,
    "total": 25,
    "as_of": "2026-09-24T00:00:00Z",
    "scope_note": "Held back in MCP server until Workers queue crawler rework completes."
  },
  "deltas": [],
  "resources": [],
  "next_actions": []
}

LLM Agent Workflow

Claude & Cursor Integration

Step 1

Prompt Trigger

"Run a health crawl of example.com."

Step 2

Agent Decision & Reasoning

Tool is built (src/domains/audit/audit-site.ts) but held back from active MCP registration (implemented: false) until asynchronous Workers queue crawling replaces synchronous loops.

Step 3

Recommended Follow-up Tools

inspect_indexing →

Provenance Guarantees & Error Handling

Reliability

Source Classcrawl

Inspection Methodself_crawl.site_health

OFE Calibration Score0.80 (OFE 1.0 scale)

Cache DurationNone (Live)

Every fact emitted by Vouched includes source_class, method, observed_at ISO timestamp, and a published confidence score. LLM agents can inspect these citations to distinguish first-party verified facts (e.g. Search Console) from modeled competitor estimates.

Billing & API Keys

Free on every plan: it reads your own Google data, so there's no per-call cost. Works the same when self-hosted.

Self-host on GitHub (MIT)Try on Cloud plan