Unblocked is Good to agents.

What we found

• —An agent looking for this API would find it effortlessly: docs and marketing robots.txt explicitly allow-list every major AI crawler (GPTBot, ClaudeBot, anthropic-ai, Google-Extended, PerplexityBot) and carry Content-Signal ai-input=yes, and a public OpenAPI 3.0.3 spec sits at a predictable path (getunblocked.com/api/v1/public-api.json).
• —An agent could go from zero to a working integration using only the docs: the quickstart chains a complete workflow (create token → create collection → add document → ask question → poll for answer) with copy-pasteable, realistically-valued examples.
• —Capability boundaries are unusually explicit — an agent knows the hard limits up front (25 collections/team, 10MB/request, 1,000 questions/day, midnight-PST quota reset, field-length constraints) instead of discovering them by failing.
• —There's an official Unblocked MCP server (listed on PulseMCP and declared in .well-known/mcp.json), so agents have a native tool-calling path, not just REST — plus every doc page is retrievable as clean markdown via Mintlify .md URLs, llms.txt, and a 364KB llms-full.txt.
• —The gaps an agent would feel: no AGENTS.md for coding-agent context, and error docs stop at 'returns 429' without telling an agent to honor Retry-After or how to recover — self-correction guidance is thin.

What to change

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

1. 01Add error-recovery guidance: for each documented error (especially 429), state the concrete recovery step (e.g., 'wait until the quota resets at midnight PST' or honor a Retry-After header) so agents self-correct instead of hard-failing — this is the biggest comprehension lever remaining.
2. 02Publish an AGENTS.md in the primary public repo (github.com/unblocked/skills) with API base URL, auth, and rate-limit context for coding agents — a 3-weight discovery gap and a quick win.
3. 03Add an API-focused llms.txt (or /api-reference/llms.txt) that indexes only the API surface rather than the current product-wide index that mixes billing, SSO, and data-source pages — sharpens token efficiency for agents that only want to call the API.
4. 04Expose and document rate-limit response headers (e.g., X-RateLimit-Remaining, Retry-After) so agents can throttle proactively rather than reactively.
5. 05Point .well-known/mcp.json at a branded production MCP URL rather than the Mintlify preview host (unblocked.main-kill-isr.mintlify.me) to signal stability to agent runtimes.