Replicate is Agent-Ready to agents.
Discry independently scored how well an AI agent can discover and understand the Replicate 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 · 90/100What we found
- Replicate achieves strong discovery through a combination of public OpenAPI spec, well-structured llms.txt with code examples, comprehensive sitemap, and official MCP server with OAuth — a near-complete agent discovery stack.
- An agent would find Replicate's Content-Signal header in robots.txt (ai-input=yes) — a forward-looking signal that explicitly invites AI consumption of content.
- An agent building integrations would benefit from extensive multi-step workflow guides (Next.js apps, Discord bots, CI/CD pipelines, SwiftUI apps) covering real use cases end-to-end.
- The OpenAPI spec at api.replicate.com/openapi.json gives agents complete programmatic access to Replicate's API schema without documentation parsing.
- The only significant gaps are llms-full.txt (absent) and .well-known/mcp.json (absent) — easy additions given the infrastructure already in place.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Add llms-full.txt with comprehensive markdown documentation of the HTTP API including all endpoints, authentication patterns, and webhook handling.
- 02Add .well-known/mcp.json pointing to the official MCP server — would formalize what already exists on PulseMCP.
- 03Expand error recovery guidance in HTTP API reference with specific actionable steps for common failure modes (model cold starts, timeout handling, webhook failures).
- 04Add explicit markdown format support (.md links or content negotiation) to documentation pages for better LLM parsing.
- 05Document idempotency patterns for prediction creation to help agents safely retry failed requests.
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.
Well-documented API with Bearer token auth. OpenAPI spec publicly accessible at api.replicate.com/openapi.json. Webhooks documented for async prediction callbacks. Rate limits and error handling covered in HTTP reference. No explicit idempotency key mechanism documented.