Apify is Agent-Ready to agents.
Discry independently scored how well an AI agent can discover and understand the Apify 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 · 100/100What we found
- An agent could go from discovery to first call almost frictionlessly: Apify ships a valid OpenAPI 3.1.2 spec (JSON + YAML), a concise llms.txt, a full 43MB llms-full.txt, an AGENTS.md, and a 'Copy for LLM' button on every docs page — one of the most complete agent-readiness footprints seen in a scan.
- Apify is explicitly agent-native: the API reference documents agentic payments (x402 + Skyfire) so an AI agent can authenticate and pay for Actor runs without a human account, and the official Apify MCP server is listed across PulseMCP, Glama, and Smithery.
- Error recovery is genuinely actionable, not just a code list: the reference gives a typed error table plus an explicit exponential-backoff pseudo-code algorithm for handling 429s — an agent can self-correct instead of guessing.
- The core run→monitor→fetch workflow is documented end-to-end in the API intro (run Actor → poll Get run → fetch Get items / Get record), so an agent chaining operations doesn't have to reverse-engineer the sequence from isolated endpoints.
- The only soft spot is llms.txt sizing: the docs-domain llms.txt is ~89KB (>50KB), large enough that a naive agent could over-fetch; the leaner API-focused llms.txt at apify.com (~14KB) is the better agent entry point but isn't the one served from the docs base domain.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Trim or split the docs.apify.com/llms.txt (~89KB) toward the leaner, API-focused ~14KB version served at apify.com/llms.txt, so agents landing on the docs domain get a right-sized, section-with-descriptions index under the ~12K-token guideline.
- 02Publish a /.well-known/mcp.json at the docs (or apify.com) root advertising the official Apify MCP server and its tool/auth declarations — the MCP server already exists and is registry-listed, so this is a near-zero-effort discovery win.
- 03Document idempotency support (or its absence) for run-creating endpoints like Run Actor; agents that retry on network errors need to know whether a duplicate POST creates a duplicate run.
- 04Surface the agentic-payments (x402/Skyfire) and 'Copy for LLM' capabilities in the llms.txt header itself, so agents discover Apify's agent-native features without scraping the full reference.
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 documented to an unusually high standard: machine-parseable JSON errors ({error:{type,message}}) with a common-error table, global (250k/min) and per-resource (60-400/s) rate limits surfaced via X-RateLimit-Limit headers, and two pagination schemes (offset/limit + exclusiveStartKey) each with response-shape tables. Notably, Apify documents agent-native auth (x402 + Skyfire agentic payments), letting agents pay for Actor runs without an account. Idempotency keys are not documented.