ManHourPlan Tools 掃描 4 個開發者平台首頁,評估 AI 智能體就緒度
I scanned 4 developer-platform homepages for AI-agent readiness (llms.txt, MCP, x402, OpenAPI)
ManHourPlan Tools 使用自建的 Agent-Ready Scan,按十項訊號檢查 4 個開發者平台首頁的 AI 智能體就緒度並評分。
AI assistants and agents read websites differently from people. Most of them don't run JavaScript, they look for files such as llms.txt, and they follow robots.txt rules written for their own user agents. Agents that do work for someone can also call an API or pay for it per call, but only if they can find it.
I built a small, free scanner that checks ten of these signals and gives a score out of 100: Agent-Ready Scan. Disclosure: it's my tool, and the page also sells a done-for-you setup. The scan itself is free, needs no account, and doesn't store results.
What it checks
| Check | Points |
|---|---|
| Page readable without JavaScript (first HTML response has real content) | 15 |
/llms.txt (llmstxt.org) |
15 |
/llms-full.txt |
5 |
robots.txt rules for AI crawlers (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-User, PerplexityBot, Google-Extended…) |
10 |
| XML sitemap | 10 |
| schema.org JSON-LD | 10 |
/.well-known/mcp.json (points agents to an MCP server) |
10 |
/.well-known/x402.json (pay-per-call APIs in USDC) |
10 |
/.well-known/api-catalog (RFC 9727) |
5 |
| OpenAPI description | 5 |
The MCP, x402, API-catalog and OpenAPI checks only matter if you offer an API or data. A brochure site can be in good shape without them.
Results on a few homepages (22 Nov 2026)
I ran the scan on these homepages on 22 November 2026. Sites change, so your numbers may differ:
| Site | Score | Missing |
|---|---|---|
| vercel.com | 74 | llms-full.txt, mcp.json, x402.json |
| supabase.com | 79 | mcp.json, x402.json |
| www.cloudflare.com | 84 | x402.json, api-catalog |
| manhourplan.com (my own site) | 100 | — |
Two things stood out:
-
The big platforms already do the basics. All three serve real HTML without JavaScript, have
llms.txt, a sitemap, JSON-LD and explicit AI-crawler rules inrobots.txt. -
Machine discovery of APIs is still rare. Only one of the three exposed
/.well-known/mcp.json, and none had an x402 catalog. Those are the files an agent would use to find a tool it can call (or pay for) without a human reading docs first.
Quick wins for your own site
- llms.txt: a short Markdown file at the root with a one-line description of the site and links to your key pages. It's the highest-value file here and takes ten minutes.
- robots.txt: decide on purpose which AI bots you allow. A common split is to allow search/assistant bots (OAI-SearchBot, ChatGPT-User, Claude-User, PerplexityBot) and block training bots (GPTBot, ClaudeBot, Google-Extended), or allow all. Either way, write the rules down.
-
Server-rendered content: if your homepage is an empty
<div id="root">until JS runs, most crawlers see nothing. -
If you have an API: publish OpenAPI, add an RFC 9727
/.well-known/api-catalogthat links to it, and if you run an MCP server, add/.well-known/mcp.jsonand list it in the official MCP Registry.
Limits
The scan only checks public files and the raw HTML. It can't tell you whether any AI assistant will cite or recommend you, and nobody can honestly promise that.
You can run it on your own site here: https://manhourplan.com/agent-ready/?ref=devto_ar (10 scans per IP per day). Agents can also call it as a free MCP tool at https://manhourplan.com/agent-ready/mcp.
來源:DEV Community · dev.to