# H/M Agent Surface Scanner

Evidence-based audit of the public machine-readable surface of a website.

Human interface: <https://hogarmas.net/agent-surface/>

Free API contract: <https://hogarmas.net/api/agent-surface-audit>

JSON Schema: <https://hogarmas.net/agent-surface/schema.json>

Paid evidence package: `POST https://agents.hogarmas.net/v1/agent-surface/report`

## What the free audit observes

The scanner groups public HTTP evidence into four layers:

1. **Access**: HTTPS, homepage reachability, `robots.txt`, named AI crawler rules and Cloudflare Content Signals.
2. **Retrieval**: sitemap, the community `llms.txt` proposal, HTML metadata, JSON-LD and optional Markdown content negotiation.
3. **Action**: OpenAPI, RFC 9727 API Catalog, site-declared MCP discovery, A2A Agent Card and site-declared agent payment terms.
4. **Trust**: RFC 9116 `security.txt`, policy links and HTTP cache validators on machine-readable artifacts.

## Free API

Request:

```http
POST /api/agent-surface-audit
Content-Type: application/json

{"url":"https://example.com"}
```

The response contains the observed target origin, summary, layer states and individual checks. It never includes fetched response bodies.

Limits:

- Public HTTP and HTTPS targets only.
- Local, private, reserved and common wildcard-DNS targets are blocked.
- Standard HTTP and HTTPS ports only.
- Free rate limits apply.
- The target URL and fetched content are not stored.
- Aggregate counts by channel, outcome and country are stored.
- A rotating anonymous request hash is used only for quota enforcement.

## Focused MCP

Connect: `https://agents.hogarmas.net/mcp/agent-surface`

Server Card: <https://agents.hogarmas.net/.well-known/mcp/agent-surface/server-card.json>

Registry package: `net.hogarmas/agent-surface`

The endpoint intentionally exposes exactly two tools:

- Free: `audit_agent_surface`
- Paid: `unlock_agent_surface_evidence_pack`

Resources: `mcp://agent-surface/server-card.json`, `mcp://agent-surface/capabilities.json`, `mcp://agent-surface/protocol.md`

Prompts: `audit-agent-surface`, `prepare-agent-surface-evidence-pack`

The paid prompt does not authorize payment. An agent must inspect the current x402 offer and obtain explicit operator authorization for the current 0.05 USDC Base mainnet transaction before calling the paid tool.

## Paid x402 evidence package

The paid resource costs `0.05 USDC` on Base mainnet (`eip155:8453`). Inspect the exact HTTP 402 payment terms before authorizing payment.

It adds:

- SHA-256 digests for fetched public resources.
- Selected response headers.
- Deterministic remediation templates.
- A SHA-256 integrity digest for the package.

The package is unsigned evidence. It is not certification, legal advice, a ranking score or cryptographic proof of the target site's identity.

## Truth boundary

This tool does not prove that an AI system indexed, cited, ranked, trusted, paid for or acted on a website. A present MCP or A2A document is a site declaration unless separately verified. Missing evidence means only that the scanner did not observe it at scan time.

`llms.txt`, site-declared MCP discovery and the payment manifest are reported as conventions, not presented as normative web standards.