replicate.com

Whether agents can crawl you, find you in the registries and searches they check, and trust what they find.

• ◐Sitemap present & freshrequired2/4

  Sitemap at https://replicate.com/sitemap.xml is valid XML with 39 entries, but the freshest <lastmod> is 2026-05-13 (97 days ago — older than 90 days)

  fix → Update <lastmod> dates as pages change so agents can tell what's current. Freshest lastmod seen: 2026-05-13.

  Resolved the sitemap from robots.txt or /sitemap.xml, validated XML, and checked lastmod freshness

  spec ↗report issue
• ◐Brand search discoverabilityrequired2/4

  Search "Replicate" cited replicate.com (https://sdks.replicate.com/resources/models/methods/search/, https://sdks.replicate.com/typescript/resources/$client/methods/search/, https://replicate.com/docs/topics/predictions/create-a-prediction, https://replicate.com/changelog/2023-10-05-api-for-listing-public-models/), but search "Replicate ai ml" did not cite your domain

  fix → Your domain doesn't surface when agents search "Replicate ai ml". Strengthen brand pages, structured data, and third-party citations (docs portals, GitHub, directories).

  Ran clean brand-name web searches and checked whether your domain is cited in the results

  report issue
• ✓MCP registry listingsrecommended4/4

  /.well-known/mcp/server-card.json serves JSON; searched "replicate.com" and "Replicate": listed in official MCP registry (1 of 3 reachable registries)

  Queried the official MCP registry, Smithery, Glama, and PulseMCP for servers matching your domain

  spec ↗report issue

Whether your robots policy, bot protection, and agent guidance actively admit AI agents instead of blocking them.

• ✓robots.txt present & parseablerequired2/2

  /robots.txt parses: 1 User-agent group(s), 1 Sitemap line(s)

  Fetched /robots.txt and validated it parses as a robots policy

  spec ↗report issue
• ✓AI crawler policyrequired5/5

  All 12 AI crawler user-agents (GPTBot, ClaudeBot, PerplexityBot, …) are allowed at / by robots.txt

  Evaluated robots.txt groups for the major AI agent user-agents (GPTBot, ClaudeBot, PerplexityBot, …)

  report issue
• ✓Content Signals directivesemerging2/2

  1 valid Content-Signal line(s) in robots.txt (e.g. `Content-Signal: search=yes, ai-train=no, ai-input=yes`)

  Looked for Content-Signal lines in robots.txt

  spec ↗report issue
• ✓Agent user-agent paritybetarequired4/4

  All three UAs get HTTP 200; bot bodies are ClaudeBot 100%, GPTBot 100% of the browser response (354303 bytes)

  Compared responses served to browser and AI-agent user-agents (informational while in beta)

  report issue
• ✗agents.mdrecommended0/3

  No agents.md: https://replicate.com/agents.md → HTTP 404, https://replicate.com/AGENTS.md → HTTP 404

  fix → Add /agents.md: what agents may do on your site, key URLs, auth, rate limits, and who to contact.

  Fetched /agents.md and checked for substantive agent guidance

  spec ↗report issue

Whether your pages carry machine-readable identity: structured data, llms.txt, clear copy an agent can quote.

• ✓Homepage states what you arerequired3/3

  Homepage clarity rated 5/5 — an agent can confidently say what this company does (model's one-sentence read: "Replicate is an AI/ML platform that allows developers to run, fine-tune, and deploy machine learning models using an API, with automatic scaling and infrastructure management.")

  An LLM read your homepage as an agent would and rated how confidently it could say what you do

  report issue
• ◐OpenGraph / social metadatarecommended1/2

  Homepage has og:description but is missing og:title, og:image (twitter:card absent)

  fix → Add og:title and og:image meta tags to your homepage; agents and link unfurlers use them as your canonical summary.

  Parsed homepage og:title / og:description / og:image and twitter:card meta tags

  spec ↗report issue
• ✓llms.txtrecommended4/4

  /llms.txt is valid llmstxt.org shape: H1 "Run AI with an API", 1 markdown link(s)

  Fetched /llms.txt and validated it against the llmstxt.org shape (H1, summary, curated links)

  spec ↗report issue
• ✗llms-full.txtbonusrecommended0/2

  GET /llms-full.txt returned HTTP 404

  fix → Add /llms-full.txt with expanded inline docs content so agents can load everything in one fetch.

  Fetched /llms-full.txt and checked for substantial inline markdown content

  spec ↗report issue
• ✗JSON-LD structured datarequired0/4

  No application/ld+json blocks found on homepage and pricing page

  fix → Embed Organization + WebSite JSON-LD on your homepage (name, url, logo, sameAs) and Product/Offer JSON-LD on pricing.

  Extracted and validated application/ld+json blocks on the homepage and pricing page

  spec ↗report issue
• ✓Markdown content negotiationrecommended3/3

  https://replicate.com/ returns markdown for `Accept: text/markdown` (content-type text/markdown)

  Requested key pages with Accept: text/markdown and probed .md twin URLs

  report issue

Whether the artifacts an agent needs to build on you — docs, API specs, SDKs, MCP servers — exist and are findable.

• ✓Developer resource discoverabilityrecommended3/3

  Search "Replicate API documentation" cited your own domain: https://replicate.com/docs/reference/http, https://sdks.replicate.com/python/resources/$client/methods/search/, https://replicate.com/docs/reference/openapi, https://replicate.com/blog/new-search-api

  Searched for your brand with developer-keyword suffixes and checked which official resources are cited

  report issue
• ✗OpenAPI spec discoverablerequired0/5

  No OpenAPI spec found: checked docs-page links and /openapi.json, /openapi.yaml, /swagger.json, /api/openapi.json, /docs/openapi.json, and api-subdomain locations

  fix → Publish your OpenAPI spec at /openapi.json and link it from docs — it's the single highest-leverage artifact for agent integration.

  Probed standard OpenAPI locations and docs links for a fetchable, parseable spec

  spec ↗report issue
• ✓Docs discoverablerequired4/4

  Docs found at https://replicate.com/docs (via homepage link, 5522 chars of readable text)

  Followed homepage nav/footer links and probed /docs, /developers, docs.{domain}

  report issue
• ✓MCP server discoveryrequired5/5

  Valid MCP server card at https://replicate.com/.well-known/mcp/server-card.json: name "com.replicate/replicate-mcp", endpoint https://mcp.replicate.com/sse

  Probed /.well-known/mcp/server-card.json, mcp.json, and docs mentions for an MCP endpoint

  spec ↗report issue
• ✓npm SDKrecommended2/2

  Official npm package "replicate" (verified via repository under the replicate GitHub org (git+https://github.com/replicate/replicate-javascript.git)): latest 1.4.0 published 2025-11-17

  Searched the npm registry for an official, domain-verified SDK package

  report issue
• ✓PyPI SDKrecommended2/2

  Official PyPI package "replicate" (verified via project_urls.homepage https://replicate.com): latest 1.0.7 published 2025-05-27

  Searched PyPI for an official, domain-verified SDK package

  report issue

Whether your specs, docs, and tools are complete and descriptive enough for an agent to use them without guessing.

• —OpenAPI validity & qualityrecommendedna

  No OpenAPI spec found — nothing to lint

  Linted the spec: descriptions, operationIds, securitySchemes, servers

  report issue
• ◐Docs qualityrecommended2/3

  Docs at https://replicate.com/docs scored clarity 4/5, completeness 3/5, runnable examples 3/5, agent-friendliness 3/5 (mean 3.3/5)

  fix → Docs scored 3/5 on completeness. The documentation provides a high-level overview of many topics but lacks deep, actionable details for an agent to perform specific tasks without further exploration.

  An LLM rated your docs for clarity, completeness, runnable examples, and agent-friendliness

  report issue
• ✗Quickstart / getting startedrecommended0/2

  No quickstart/getting-started link or heading found on docs page https://replicate.com/docs (searched 190 links)

  fix → Add a copy-pasteable quickstart (auth → first API call) — agents follow it literally.

  Looked for a quickstart/getting-started guide containing code blocks

  report issue
• —MCP tool qualityrecommendedna

  MCP endpoint is auth-gated — tools could not be listed unauthenticated

  Linted listed tools for descriptions, typed input schemas, and naming

  report issue

Response hygiene, TLS and redirect discipline, and security contact channels agents depend on at runtime.

• ✓TLS & redirect hygienerequired2/2

  http://replicate.com/ upgrades to https in 1 hop(s), 1 redirect(s) total, final HTTP 200; HSTS present

  Checked http→https redirect behavior, chain length, and TLS health

  report issue
• ✓Response speed & weightrequired1/1

  Homepage TTFB 324ms, payload 345KB, Content-Encoding: br

  Measured homepage TTFB, payload size, and compression

  report issue
• ✗security.txtrecommended0/1

  https://replicate.com/.well-known/security.txt returned HTTP 404

  fix → Publish RFC 9116 /.well-known/security.txt with a Contact and a future Expires.

  Fetched /.well-known/security.txt and validated Contact + Expires

  spec ↗report issue

Whether agents can discover your auth model machine-readably (OAuth metadata) and follow documented steps to credentials.

• ✗OAuth authorization server metadatarecommended0/3

  Neither /.well-known/oauth-authorization-server nor /.well-known/openid-configuration resolved (HTTP 404 / HTTP 404)

  fix → Serve RFC 8414 metadata at /.well-known/oauth-authorization-server with PKCE (S256) and a registration_endpoint so agents can self-onboard.

  Fetched /.well-known/oauth-authorization-server (and openid-configuration fallback)

  spec ↗report issue
• ✗OAuth protected resource metadatarecommended0/2

  No RFC 9728 metadata: https://replicate.com/.well-known/oauth-protected-resource → HTTP 404

  fix → Serve RFC 9728 metadata at /.well-known/oauth-protected-resource naming your authorization servers so agents can discover how to authenticate.

  Fetched /.well-known/oauth-protected-resource

  spec ↗report issue
• ✓Auth documentationrecommended2/2

  Auth documentation found at https://replicate.com/docs/topics/security/api-tokens (linked from https://replicate.com/docs) with code examples

  Looked for /auth.md or an authentication docs page with code examples

  report issue

Whether pricing is discoverable and machine-readable, and whether you support agent payment protocols.

• ✓Pricing discoverablerequired5/5

  Pricing page found at https://replicate.com/pricing (via homepage link) with legible price signals ("$0.015 / thousand output token").

  Followed nav/footer links and probed /pricing for a page with legible price signals

  report issue
• !Machine-readable pricingrecommendederror

  Inference call failed — check not evaluated

  Looked for Offer JSON-LD, then had an LLM attempt structured extraction of your plans

  spec ↗report issue
• ✗x402 payment supportbonusrecommended0/2

  /.well-known/x402 returned HTTP 404. No x402 support detected (bonus check — absence costs nothing).

  fix → Support x402: serve payment requirements at /.well-known/x402, or answer unauthenticated API calls with HTTP 402 plus an x402 payment-requirements payload (x402Version, accepts[]).

  Probed /.well-known/x402 and API endpoints for HTTP 402 payment-required envelopes

  spec ↗report issue
• ✗AP2 readinessemerging0/1

  No AP2 hints found — scanned llms.txt, docs page (https://replicate.com/docs); /.well-known/ap2 returned HTTP 404.

  fix → Adopt AP2 to accept delegated agent payments — see https://ap2-protocol.org.

  Scanned fetched artifacts and well-known paths for AP2 hints

  spec ↗report issue
• ✗Agentic Commerce Protocolemerging0/1

  No Agentic Commerce Protocol hints found — scanned llms.txt, docs page (https://replicate.com/docs); /.well-known/acp returned HTTP 404.

  fix → Adopt the Agentic Commerce Protocol so agent checkouts can complete against your store — see https://developers.openai.com/commerce.

  Scanned docs and specs for agentic checkout endpoints

  spec ↗report issue
• ✗Other agent payment protocolsbonusemerging0/1

  No UCP/MPP hints found — scanned llms.txt, docs page (https://replicate.com/docs).

  fix → Track emerging agent payment protocols (UCP, MPP) and adopt the ones your buyers' agents use.

  Scanned fetched artifacts for UCP/MPP protocol hints

  report issue

Whether an end user's agent can operate on their behalf: working MCP tools, published skills, agent configs.

• —Agent skill publishedbonusemergingna

  Requires the analysis phase — not yet evaluated

  Checked /skill.md, /.well-known/skills/, and skills.sh for published agent skills

  report issue
• ✗MCP handshake & tools listrequired0/4

  MCP endpoint https://mcp.replicate.com/sse returned 401/403 without resolvable resource_metadata in WWW-Authenticate

  fix → MCP endpoint https://mcp.replicate.com/sse failed handshake: auth-gated without RFC 9728 metadata. If auth-gated, return resource_metadata in WWW-Authenticate so agents can discover how to authenticate.

  Performed a streamable-HTTP initialize + tools/list against the MCP endpoint

  spec ↗report issue
• —Agent configs in public repobonusemergingna

  Requires the analysis phase — not yet evaluated

  Checked your public GitHub org's main repos for AGENTS.md / .claude / .cursor configs

  report issue

Whether the site itself is legible to non-rendering and browser agents: semantic HTML, no JS walls, accessibility.

• ✗NLWeb endpointemerging0/1

  No NLWeb endpoint detected: /.well-known/nlweb.json returned HTTP 404 (text/html); GET /ask?query=hello returned HTTP 404 (text/html)

  fix → Consider exposing an NLWeb /ask endpoint (and /.well-known/nlweb.json) for conversational access to your content.

  Probed /.well-known/nlweb.json and the /ask endpoint

  spec ↗report issue
• ◐Semantic HTML structurerequired1.5/3

  3/6 semantic signals present on the homepage (missing: <main> landmark, no heading-level skips, text-to-markup ratio ≥ 0.10)

  fix → Add the missing structure: <main> landmark, no heading-level skips, text-to-markup ratio ≥ 0.10. Agents parse structure, not pixels.

  Scored landmark elements, heading hierarchy, and text-to-markup ratio on the homepage

  report issue
• ✓Content readable without JavaScriptrequired2/2

  Homepage raw HTML contains 12158 chars of visible text without JavaScript

  Measured visible text in the raw, unrendered homepage HTML

  report issue
• ✓Accessibility basicsrequired2/2

  Homepage accessibility: 4/5 checks passed — failing: labeled inputs (1/8 unlabeled)

  Static checks: lang attribute, title, alt coverage, labeled inputs, landmarks

  report issue
• ✗WebMCPemerging0/1

  No WebMCP evidence on homepage: no application/webmcp script and no navigator.modelContext reference

  fix → Consider WebMCP to expose page actions as tools to browser agents.

  Looked for WebMCP script declarations on the homepage

  spec ↗report issue