ClickUp is Good to agents.
Discry independently scored how well an AI agent can discover and understand the ClickUp 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 · 76/100Comprehension
55% of score · 78/100What we found
- An agent landing on the docs is explicitly invited in: every page carries a 'For AI agents: visit /llms.txt' banner and a top-of-page pointer to the Markdown documentation index — ClickUp is one of the few APIs actively signposting agents.
- Clean Markdown is available for the entire site: any doc or reference page can be fetched as `.md` (e.g. /reference/createtask.md), and each reference page embeds the full OpenAPI definition for that operation — near-ideal for LLM consumption.
- Both OpenAPI specs are public and live: V2 (508KB JSON) and V3 (131KB) both return 200 with import instructions for Postman — an agent can generate a typed client with zero scraping.
- The biggest discovery gap is AGENTS.md: no AGENTS.md exists in ClickUp's GitHub org, so a coding agent working in a repo gets no first-party setup context (the community fills this with third-party MCP servers and CLIs instead).
- Comprehension is solid but uneven: rendered guide pages (Get Tasks, OAuth flow) are richly task-oriented, while raw OpenAPI summaries ('Create a new task.') and many parameter descriptions are terse or empty — an agent gets good context from guides but thin context from the spec alone.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Publish an AGENTS.md in the primary public GitHub repo (and link it from the docs) so coding agents get first-party auth, base-URL, and rate-limit context — the single highest-leverage discovery fix from a current fail.
- 02Add a llms-full.txt (ReadMe supports it) so an agent can ingest core API capabilities in one fetch instead of following the llms.txt index page-by-page — this would lift token efficiency from partial to pass.
- 03Enrich OpenAPI-level descriptions: give every endpoint a one-line task-oriented summary ('Create a task in a List, optionally with assignees, custom fields, and due dates') and fill the many empty parameter `description` fields, so the spec is self-sufficient for agents that read it directly.
- 04Make error recovery explicit: on the Common Errors and Rate Limits pages, add an actionable 'how to fix' step for the top errors (e.g. 'On 429, back off using the Retry-After / X-RateLimit-Reset header') rather than only stating the cause.
- 05Serve a .well-known/mcp.json pointing at ClickUp's official MCP server so agents can auto-discover the hosted MCP without reading a setup doc.
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 (personal token + OAuth2), rate limits (per-plan tiers, 429 handling), and pagination (100/page, dedicated comments-pagination guide) are all clearly documented. Errors are returned as JSON with named codes and a dedicated Common Errors page. Idempotency keys are not documented.