---
name: agent-ready
description: Make a website ready for AI agents (ChatGPT, Claude, Perplexity, Gemini) so they can read it, recommend it and finish tasks on it - buy, sign up or get an API key. Use when the user asks to make their site agent-ready, agent-friendly or AI-ready, improve AEO or "SEO for AI", add llms.txt, unblock ChatGPT or Claude, add structured data for AI, or raise their AgentsReady score. Scores the site with agentsready.dev, then fixes each gap in the codebase and verifies it.
license: MIT
metadata:
  author: AgentsReady (agentsready.dev)
  version: "1.0.0"
---

# Make a site agent-ready

An agent-ready site is one an AI agent can fetch without a bot wall, read in plain HTML, understand from structured data, and use to finish the job a person sent it to do. AgentsReady scores this out of 100: **60 core points** every site needs, plus **40 for the job** agents should do there (`recommend`, `buy`, `signup` or `api`).

Work in this order: measure, fix the biggest gaps first, verify each one, measure again.

## 1. Measure

Find the production domain (ask only if it isn't in the repo: check `package.json` homepage, `next.config.*`, `astro.config.*`, `vercel.json`, `netlify.toml`, CNAME, README). Then get the score:

- **MCP (preferred):** if the `agentsready` MCP server is connected, call `check_site` with `{ "domain": "example.com" }`. Add `"job"` if you know what agents should do on the site.
- **HTTP:** `bash scripts/check.sh example.com` or `curl -s "https://agentsready.dev/api/check?domain=example.com"`.

The result has `score`, `job`, `fixes` (biggest first, each with `check`, `points`, `fix`) and `passed`. Show the user the score, the detected job and the top 5 fixes before changing anything. If the detected job is wrong (a shop scored as `recommend`, say), re-run with the right `job`.

## 2. Find the stack

Look at the repo to see what serves the site and where each fix goes. See [references/stacks.md](references/stacks.md) for Next.js, Astro, Nuxt, SvelteKit, plain HTML, WordPress, Shopify, Webflow, Framer and docs platforms (Mintlify, GitBook, Docusaurus, Fern). If the site isn't in this repo (hosted builder, separate repo), say what to change and where, and stop.

## 3. Fix, biggest gap first

For each fix in the result, apply the change from [references/fixes.md](references/fixes.md), then run its verification command against a local or preview build. Rules:

- **Never weaken security.** Bot walls, WAF rules and rate limits: explain the change and ask before touching them. Allow user-triggered agents (ChatGPT-User, Claude-User, Perplexity-User); don't open the site to everything.
- **Facts only.** Structured data, llms.txt, FAQ answers and "as of" dates must match what the site actually says. Never invent prices, ratings, addresses or reviews.
- **Small, reviewable changes.** One commit per fix area (access, readability, structured data, llms.txt, job path).
- Skip fixes that don't apply (no checkout on a docs site) and say why.

## 4. Verify and re-score

After deploying (or on a preview URL), score again with `check_site` or `scripts/check.sh`. Report before and after, what changed, and what's left that needs a human: bot rules in a CDN dashboard, CAPTCHAs on sign-up, a sales-only API, KYC.

For the gaps a scan can't see (email verification, phone OTP, card before trial), suggest the free agent run at `https://agentsready.dev/?check=<domain>`, which sends a real agent to try the job and names every human gate.

## Reference

- [references/fixes.md](references/fixes.md): every check, what passes, the change and how to verify it
- [references/stacks.md](references/stacks.md): where each fix lives per framework or platform
- Live standard: `get_standard` (MCP) or https://agentsready.dev/standard
- Full checklist (50 items): `get_checklist` (MCP) or https://agentsready.dev/checklist