amentum.space is Poor to agents.

What we found

• —An agent that reaches the service can consume it cleanly: every endpoint publishes a valid OpenAPI 3.0.0 spec at a stable URL (e.g. ocean.amentum.io/openapi.json) with task-oriented summaries, enumerated parameter values, realistic curl samples, and example JSON responses — the strongest signal in this scan.
• —Discovery is the failure mode, not comprehension. The scanned domain amentum.space is a marketing redirect to amentum.io with no robots.txt, no llms.txt, and no sitemap, so an agent starting from the brand domain has no machine-readable trail to the per-service API subdomains where the real docs live.
• —Amentum ships a first-party hosted MCP server (mcp.amentum.io/mcp/{service}/) with a documented client config and example agent tasks — genuinely ahead of most peers — yet it is absent from Glama, Smithery, and PulseMCP, so agents discovering tools through registries will never find it.
• —Naming and structure are model-friendly: strict snake_case fields, a uniform {value, units} response envelope, consistent lat/lon/depth/date parameters, and answer-first ReDoc pages that lead with method, URL, and purpose.
• —Error and boundary documentation is descriptive but not prescriptive: 408 messages explain likely causes ("salinity dataset was not ready... will be available soon") and parameter bounds are enumerated, but there is no explicit recovery guidance, quota numbers, or dedicated limitations page for an agent to self-correct against.

What to change

Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.

1. 01Add a robots.txt and an llms.txt at amentum.space (or redirect it cleanly to amentum.io) that names the API and links the per-service OpenAPI specs and the /mcp page — this is the single highest-impact fix and would lift Discovery from F on its own.
2. 02Publish a sitemap.xml on amentum.io covering the per-service doc pages (ocean, atmosphere, avrad, globalmagnet, gravity) so crawlers and agents can enumerate the documentation surface.
3. 03List the hosted MCP server in the major registries (Glama, Smithery, PulseMCP) and add a /.well-known/mcp.json so agents discover the tool declarations you already expose.
4. 04Add multi-language code samples (Python/JS alongside curl) to the OpenAPI x-codeSamples, and add 2-3 end-to-end workflow guides (e.g. the marine-routing chaining shown on the MCP page) inside the docs rather than only on marketing pages.
5. 05Quantify capability boundaries in the docs — publish rate-limit/quota numbers and header names, and give each documented error an explicit recovery step (e.g. retry-after guidance for 408, valid-range hints for out-of-bounds coordinates).