Enable Banking is Poor to agents.

What we found

• —An agent trying to DISCOVER this API programmatically hits a wall: there is no robots.txt (HTTP 404), no llms.txt, no llms-full.txt, no .well-known/mcp.json, and no AGENTS.md in the GitHub org — none of the machine-readable discovery surfaces an agent would probe first exist.
• —Once an agent REACHES the docs, comprehension is strong: the reference is a cleanly structured single page (H1 sections -> H2 endpoints -> H3 Parameters/Responses) where every endpoint leads with its method and path, uses consistent snake_case, and documents a standardized ErrorResponse JSON with named, described error codes.
• —The Quick Start reads like an agent runbook: it chains the full account-information flow (list ASPSPs -> start auth -> authorize session -> fetch balances -> fetch transactions) and the full payment flow end-to-end, with realistic, copy-pasteable Python AND JavaScript examples using plausible IBANs, amounts, and dates.
• —No downloadable OpenAPI/Swagger spec is exposed for the main aggregation API (api.enablebanking.com) — the reference is HTML-rendered from an underlying spec but an agent cannot fetch openapi.json/yaml. A separate eidas_broker microservice repo publishes an openapi.json, and a Jentic auto-import issue exists, but neither gives an agent the core API contract as structured data.
• —Capability boundaries are unusually well documented for this category: a dedicated Limitations section (Mock ASPSP has no payment initiation; transactions batch in 10s; sandbox filtering unsupported except Nordea/Handelsbanken) plus explicit rate-limit and sandbox-instability caveats mean an agent learns key constraints from docs rather than by failing.

What to change

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

1. 01Add an llms.txt at the domain root pointing to the Quick Start, API reference, sandbox, and error-code sections — this is the single highest-leverage discovery fix and would let agents orient in under 5,000 tokens instead of parsing a 196KB HTML reference.
2. 02Publish a downloadable OpenAPI 3.x spec (openapi.json / openapi.yaml) for the main aggregation API at a stable, linked path in the reference — the doc is already spec-generated, so exposing the source contract turns openApiSpec from partial to pass and unlocks automatic client/tool generation.
3. 03Add a robots.txt that explicitly allows GPTBot, ClaudeBot, Google-Extended, and CCBot (and re-list the docs sitemap) — right now the 404 gives crawling agents no signal at all.
4. 04Upgrade error documentation from 'what went wrong' to 'what to do next': add Retry-After / rate-limit-header names and per-error recovery guidance (e.g. what to do on EXPIRED_SESSION, DATE_FROM_IN_FUTURE, ASPSP_RATE_LIMIT_EXCEEDED) so an agent can self-correct instead of stalling.
5. 05Add an AGENTS.md to the enablebanking-api-samples repo summarizing the auth model, base URL, core flows, and gotchas, and publish or endorse a first-party MCP server so the API appears in Glama/Smithery/PulseMCP under its own name rather than only inside a third-party wrapper.