Temporal is Agent-Ready to agents.
Discry independently scored how well an AI agent can discover and understand the Temporal 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 · 90/100Comprehension
55% of score · 96/100What we found
- An agent landing on docs.temporal.io can fetch any page as clean raw Markdown by appending `.md` to the URL, and has both an llms.txt index (123KB, with descriptions) and a full llms-full.txt corpus (5.3MB) — a top-tier markdown-availability story that almost no API in the directory matches.
- An agent looking for end-to-end, agent-relevant guidance finds a dedicated `/ai-cookbook` with runnable agentic-loop, tool-call, human-in-the-loop, and durable-MCP-server recipes for Claude and OpenAI — Temporal documents how to BUILD agents, not just how to call an API.
- An agent needing self-correction logic finds genuinely actionable error-recovery docs: a best-practices error-handling page with explicit retryable-vs-non-retryable decision rules, idempotence design, and per-language (Python/Go/.NET/Ruby/Java/TS) error guides — far beyond a bare error-code table.
- The only meaningful discovery gap is `.well-known/mcp.json` (404). Temporal is covered by community MCP servers in PulseMCP and Glama (Mocksi Temporal Workflows, stevekinney/temporal-mcp), but there is no first-party MCP manifest at the documented domain.
- A nuance an agent must navigate: the public OpenAPI/HTTP spec (saas-api.tmprl.cloud) covers only the Cloud Ops control plane (namespaces, users, keys) — individual Workflows and Activities are driven through language SDKs over gRPC, not a REST surface.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Publish a first-party `/.well-known/mcp.json` (or a docs-linked official MCP server) so agents can discover Temporal tooling without relying on third-party PulseMCP/Glama listings — the single easiest discovery win remaining.
- 02Trim or split llms.txt: at 123KB it exceeds the ~50KB agent-friendly budget. Offer a lean, API-surface-focused index alongside the comprehensive one so token-constrained agents can orient in a single fetch.
- 03Add a concise top-level 'core capabilities in 5 minutes' overview (workflows, activities, workers, task queues, signals/queries) to reduce the tokens an agent needs to grasp the durable-execution model before diving into per-language guides.
- 04Make the control-plane vs. execution-plane distinction explicit at the top of the Cloud Ops API page (e.g., a callout: 'This API manages account resources — to run Workflows, use an SDK') so agents don't mistakenly expect a REST endpoint for Workflow execution.
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.
Execution characteristics are thoroughly documented: API key + mTLS auth for the API (plus SAML SSO and Service Accounts for control-plane access), a gRPC error model with explicit retryable/non-retryable Application Failure semantics, per-Namespace rate limits (Actions Per Second / Operations Per Second with named metrics), cursor-based pagination via next_page_token on List APIs, and extensive idempotency guidance (idempotent Activity design plus Workflow ID reuse policies).