JetBrains is Needs Work to agents.
Discry independently scored how well an AI agent can discover and understand the JetBrains API from what’s public — not whether it’s usable. Below: every signal we checked, what’s costing the score, and what to change.
SCORED UNDER RUBRIC 1.2 · A full re-launch under Discry Score 2.5 — a new behavioral instrument, not comparable to these scores — is in progress.
Discovery
45% of score · 67/100Comprehension
55% of score · 78/100What we found
- An agent can fully self-serve the primary JetBrains API: a valid OpenAPI 3.0.1 spec for the YouTrack REST API (v2026.3) is publicly reachable at youtrack.jetbrains.com/api/openapi.json, and robots.txt allows all AI crawlers — strong machine-discoverable footing.
- Endpoint pages are genuinely agent-usable: each leads with the resource, HTTP method, and a task-oriented purpose ('Read a List of Issues', 'Add a New Issue'), followed by realistic JSON request/response samples with plausible values — an agent can copy-paste and adapt immediately.
- A dedicated REST API Troubleshooting section is a standout: it documents real errors in an Error Message -> Cause -> Solution format with corrected request bodies (e.g., fixing a 500 by swapping StateProjectCustomField for StateIssueCustomField), telling an agent exactly what to DO on failure.
- The biggest discovery gap: no llms.txt, llms-full.txt, or .well-known/mcp.json. An agent optimizing for LLM-native ingestion finds nothing and must fall back to HTML-to-markdown scraping (which converts cleanly, but isn't offered natively).
- Capability boundaries are only partially exposed: per-endpoint permissions, required fields, read-only markers, and a 42-item pagination default are documented, but there is no rate-limit documentation or idempotency guidance, so an agent would discover request throttling only by hitting it.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Publish an llms.txt (and ideally llms-full.txt) at jetbrains.com covering the YouTrack/Hub REST API surface — a curated, API-focused index of auth, endpoints, and the OpenAPI spec. This is the single highest-impact discovery fix and would lift the two currently-failing llms.txt checks.
- 02Add explicit rate-limit documentation (limits, retry-after / rate-limit header names, and recommended backoff) to the Get Started section. Agents chaining many calls need this to self-throttle rather than fail.
- 03Serve documentation as native markdown via content negotiation (Accept: text/markdown) or per-page .md endpoints. The HTML already converts cleanly, so this is a low-cost win for token-efficient agent consumption.
- 04Expand the realistic examples beyond curl/HTTP to include at least one additional language (Python or JavaScript) per core endpoint, so SDK-oriented agents get copy-pasteable code, not just raw requests.
- 05Publish a .well-known/mcp.json manifest (or link the official/community YouTrack MCP servers from the docs) so MCP-capable agents can discover the tool surface directly instead of via third-party registries.
Execution coverage · INFORMATIONAL, UNSCORED
Whether an agent can actually complete a call and recover from errors is the deeper Audit layer — documented here, but not part of the Discry Score.
Auth is well documented (permanent Bearer tokens with scoped access, plus OAuth 2.0). Errors return machine-parseable JSON (error, error_description) and a dedicated troubleshooting section shows recovery. Pagination is documented for both $top/$skip and activity cursors. No explicit rate-limit documentation (only a default 42-item collection cap), and idempotency keys are not mentioned.