DeepSeek is Needs Work to agents.

What we found

• —An agent trying to *discover* DeepSeek programmatically finds almost nothing: no robots.txt, no llms.txt, no llms-full.txt, no .well-known/mcp.json, and no official AGENTS.md — the docs site returns 404 for every agent-discovery convention.
• —Once an agent reaches the docs, comprehension is excellent: endpoints are task-described, code examples span 9 languages with realistic values, and OpenAI/Anthropic compatibility means agents can reuse existing SDKs with near-zero new learning.
• —Error recovery is a standout — the Error Codes page pairs every HTTP status with an explicit cause and fix (e.g. 429 → 'pace your requests'; 402 → top-up link), and the Rate Limit page tells clients how to handle keep-alive lines and 429s.
• —No official OpenAPI spec is published for download, but the reference is clearly generated from one (Docusaurus OpenAPI) and first-party SDKs are auto-generated from a spec; community mirrors and OpenAI-compat specs fill the gap partially.
• —DeepSeek is well-represented in MCP registries (Glama, PulseMCP, Smithery) — but every listing is community-built, not an official DeepSeek server, so an agent can't rely on a vendor-maintained tool surface.

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 (and llms-full.txt) at api-docs.deepseek.com covering the core chat-completion flow, thinking mode, tool calls, and error handling — this is the single highest-impact discovery fix and Docusaurus plugins generate it automatically.
2. 02Add a robots.txt that explicitly allows AI crawlers (GPTBot, ClaudeBot, CCBot, Google-Extended) — right now the missing file reads as 'no signal' and blocks nothing intentionally but earns no credit.
3. 03Expose the OpenAPI spec that already drives the reference docs at a stable public URL (e.g. /openapi.json) so agents and codegen tools can consume it directly instead of scraping the interactive explorer.
4. 04Publish an official DeepSeek MCP server and register it on Glama/PulseMCP/Smithery so agents get a vendor-maintained tool surface rather than relying on ~8 community forks of varying quality.
5. 05Add an AGENTS.md to the primary deepseek-ai coding repos with setup/usage context for coding agents — a rising standard the org's own agent tooling already generates for downstream users.