Report card · Coding agents
Replit
69/100. Sells agents. Hasn't met one.
Scanned https://replit.com/ on October 30, 2026.
Subjects
- Arobots.txt rules for AI agentsGood work.All 11 AI agents checked may fetch every key page. No AI agent is named, so they all fall back to the * group.Extra credit: Add explicit User-agent groups for the AI agents you want, so your policy is a decision rather than a default.
- Allms.txt and llms-full.txtGood work./llms.txt found (20191 chars, 229 link(s)). /llms-full.txt found.
- ASitemapGood work.Sitemap index at https://replit.com/sitemap.xml (from robots.txt) with 5 entries.
- AServer-rendered readable HTMLGood work.All 1 key page(s) return readable HTML without running JavaScript.
- FMarkdown versions of key pagesHands in every page as HTML soup. Markdown was on the syllabus.1 of 1 key page(s) have no markdown version: /.Extra credit: Serve markdown for key pages, either at the same path plus .md or when the request sends Accept: text/markdown, and advertise it with <link rel="alternate" type="text/markdown" href="...">. Markdown cuts token cost and removes layout noise for agents.
- AStructured data (JSON-LD)Good work.Every key page has valid JSON-LD.
- FOpenAPI and API docs discoverabilityThe API docs are hiding. Agents shouldn't need a treasure map.Key pages link to API docs (https://docs.replit.com/extensions/api/auth.md), but no machine-readable spec was found at /openapi.json, /openapi.yaml, /swagger.json, /api/openapi.json, /.well-known/api-catalog.Extra credit: Publish an OpenAPI spec at /openapi.json and link it from your API docs and llms.txt. Agents can call an API they can read a spec for. If the product has no API, disable this check in config ("api-docs": { "enabled": false }).
- DMCP server discoverabilityMentions MCP, never shows up with it.An MCP endpoint is documented (https://docs.replit.com/tutorials/mcp), but no /.well-known MCP discovery document.Extra credit: Publish /.well-known/mcp.json describing the server. Discovery through /.well-known/mcp.json is still a draft proposal, so also document the server URL, transport, and auth in llms.txt and your docs.
- FAuth, signup, and bot-challenge frictionEverything worth reading is behind a login wall.1 of 1 key page(s) put friction in front of an agent: / (bot challenge page).Extra credit: Keep docs, pricing, and the product overview public. Let verified AI agents past bot challenges (most WAFs allowlist known agent user agents), and offer API keys or OAuth that an agent can use without a browser CAPTCHA.
- FPage weight and token costThis homepage weighs more than my car. Agents pay by the token.1 of 1 key page(s) cost over 20k tokens for an agent to read: /.Extra credit: Cut the token cost of /: serve a markdown version, move inline scripts, styles, and SVG out of the HTML, and avoid large inlined JSON state.
Think the grade is wrong? Fix the site and ask for a regrade, or run the scanner yourself. Scores are reproducible. How grading works.