Mondu is Needs Work to agents.

What we found

• —An agent could self-serve a working integration: the Getting Started reference lays out environments, Api-Token auth with a copy-pasteable curl call, a structured JSON error format (field-level name/details), explicit rate limits (500 req/s), and a 429 -> exponential-backoff recovery instruction all on one concise page.
• —Discovery is only half-built. robots.txt cleanly allows all AI crawlers, a well-structured API-focused llms.txt exists, and every reference page embeds its full OpenAPI 3.0.1 definition inline as fetchable markdown — but there is no consolidated spec at a standard path, no llms-full.txt, no .well-known/mcp.json, no AGENTS.md, and no MCP registry listing, so an agent must assemble the spec page-by-page.
• —Endpoint docs are genuinely task-oriented: pages lead with a 'Use Case' walkthrough (e.g. Create Order explains the full checkout->authorize->confirm->ship lifecycle and the 90-day order validity) rather than dumping the raw interface — an agent learns what an endpoint is FOR, not just its shape.
• —Multi-step workflows are well covered — 'Start integrating' is a numbered end-to-end guide (credentials -> order creation -> confirmation -> shipment), backed by integration-best-practices, marketplace setup, and order-state overviews — so an agent can chain operations for real tasks.
• —A casing mismatch would trip an agent mid-loop: request/response schemas are snake_case (external_reference_id, tax_cents) but the error-response field paths are camelCase (lines.0.lineItems.1.externalReferenceId), so an agent parsing a 422 to correct a field can't map the error path back to the field it sent.

What to change

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

1. 01Publish a single consolidated OpenAPI spec at a conventional, discoverable path (e.g. /openapi.json or a link in llms.txt). The full 3.0.1 definition already exists embedded per-page — exposing it as one downloadable file would flip the highest-weight discovery check (openApiSpec, weight 5) from partial to pass.
2. 02Fix the casing inconsistency between schema fields (snake_case) and error-response field paths (camelCase). Emitting error 'name' values that exactly match request field names lets an agent auto-correct a rejected payload without guessing.
3. 03Add an llms-full.txt (concatenated markdown of the core API guides + reference) — the docs already serve clean .md per page, so generating a single full-text file is low-effort and adds a discovery signal agents increasingly expect.
4. 04List the Mondu API in at least one MCP registry (Glama/Smithery/PulseMCP) or ship a lightweight MCP server; nothing surfaces today, so an agent discovering tools via those registries won't find Mondu.
5. 05Add multi-language code samples (Python/Node/PHP) beside the existing curl + Postman collection on endpoint pages, and lead endpoint pages with method + URL before the Use Case prose, to solidify the realistic-examples and answer-first signals.