Agentic Discovery Audit: A 1‑Page Checklist to Make Your App Discoverable to AI Agents

Claiming your app is discoverable requires three concrete properties: (1) canonical metadata that describes the app and its capabilities (who you are and what you do), (2) machine-readable entry points that an agent can call (APIs, webhooks, or action entryPoints), and (3) commerce / pricing hooks when the capability is billable (product feeds, SKU metadata, or pricing endpoints).

For auditability, each property must be verifiable with a single URL or JSON example and a test response. That’s the heart of this one‑page checklist: every row is a provable artifact an engineer or agent can fetch and validate.

Design the page as a compact table or checklist where each line is a single auditable claim. Keep columns: Claim ID, Claim (what to assert), Proof URL / JSON, Example response (snippet), Pass/Fail, Notes. Each line must be satisfiable with one GET request and one 1–3 line sample response.

Below are the rows you need. You can convert this into a CSV, Google Sheet, or a single Markdown page in your repo that an auditor or partner can fetch.

Don’t ask agents to infer. Provide short, copy‑pasteable examples: a minimal SoftwareApplication JSON‑LD, a sample ACP/UCP manifest snippet (capability name + entryPoint), and a pricing endpoint response. Keep examples under ~60 lines each so they’re human‑auditable.

The audit should mandate both a successful and an error sample for each endpoint (e.g., 200 success with content, 401 unauthorized). That makes troubleshooting faster for integrators and agent implementers.

Validation should be three lines: (A) fetch metadata JSON‑LD and assert @type and name, (B) fetch manifest and assert presence of declared capability and entryPoint.urlTemplate, (C) call pricing endpoint with example SKU and assert price/currency fields. Use curl or Postman snippets on the single audit page.

Automate where possible. Add a small CI job (GitHub Actions or similar) that runs the three checks and posts a pass/fail badge on your repo or product page. The single audit page is much more convincing when it includes a live badge that proves the endpoints still return the expected fields.

Security and auth matter. Agents will prefer standardized auth flows (OAuth 2.0 + scopes) and clear rate limits. For paid capabilities, include billingModel metadata (one_time, subscription, per_call) and link to terms and refund policy in machine-readable fields where possible.

Think like a consumer of capabilities: minimize required consent steps, give a sandbox SKU or free tier for agent testing, and keep ids stable (immutable SKU and capability ids). That stability is what allows directories and agents to cache and reuse capability descriptions safely.