codat.io is Needs Work to agents.

What we found

• —An agent that reaches the docs can complete real tasks end-to-end: the QBO Bank Feeds tutorial chains create-company to create-connection to create-source-account to generate-credentials to create-bank-transactions, each step task-framed, with realistic values (amounts like 2088.76, real UUIDs, 'Elaborate Events, Inc') and code in C#, TypeScript, Python, and Go.
• —An agent trying to DISCOVER Codat programmatically hits a wall: there is no llms.txt, no llms-full.txt, no .well-known/mcp.json, no AGENTS.md in the codatio repos, and no Codat MCP server in any major registry (Glama, Smithery, PulseMCP) — every agent-native discovery signal fails.
• —A well-maintained, publicly accessible OpenAPI spec set (github.com/codatio/oas, updated within the last week) covers every product in both JSON and YAML, so an agent that can find it gets a complete, current machine-readable contract.
• —Error and boundary handling is genuinely agent-friendly: error responses are parseable JSON with correlationId, each status code has explicit recovery guidance (402 tells you to delete companies and retry, 429 points to Retry-After and rate-limit docs), and the tutorials call out hard limits inline (max 1000 transactions/sync, no future-dated writes, chronological ordering).
• —Scraped markdown is polluted by glossary tooltips — every occurrence of terms like 'API', 'company', or 'sync' expands into a full inline definition, so an agent ingesting a page pays a repeated token tax and must filter noise; combined with the absence of any llms.txt or overview digest, understanding the multi-product surface costs more tokens than it should.

What to change

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

1. 01Publish an llms.txt at docs.codat.io/llms.txt with a titled, described index of the core product APIs (Platform, Bank Feeds, Bill Pay, Expenses, Lending) and links to the key using-the-api pages — this is the single highest-impact discovery fix and would lift both llmsTxt and llmsTxtQuality (5 discovery points).
2. 02Add an AGENTS.md to the primary public repos (codatio/codat-docs and the client-sdk-* repos) with build, auth, and usage context for coding agents — a rising standard now weighted 3 in the rubric and currently a total miss.
3. 03Suppress or strip the glossary-tooltip term expansions from the markdown/scraped output (or serve a content-negotiated clean-markdown version), so agent ingestion is not taxed by repeated inline definitions — this would move markdownAvailable and tokenEfficiency from partial to pass.
4. 04List an official Codat MCP server (or a reference wrapper over the OpenAPI spec) in at least one major registry (Glama, Smithery, or PulseMCP) so agents can discover Codat through the emerging MCP ecosystem.
5. 05Add a concise, single-page capabilities overview or llms-full.txt digest so an agent can grasp the full multi-product surface in under 5,000 tokens rather than crawling several separate product API landing pages.