Site index for AI agents: /llms.txt. Documentation pages under /docs are also served as markdown at the same URL plus .md (e.g. /docs/geo-audit.md) or via an Accept: text/markdown request.

Your GEO audit, inside your codebase.

airpulse-site-review lets your coding agent review your website's code, page by page, for what AI search engines need. It reads your AirPulse audit through the MCP, confirms each finding in your source, and proposes fixes as diffs in the right files.

One command. Any agent.

Works in Claude Code, Codex, Cursor and the other agents the skills installer supports.
Installs into your repository or your user folder, and updates with npx skills update.
Needs the AirPulse MCP connected in the same agent.
Install
npx skills add airmeet/airpulse-skills
Then ask your agent

Review this site for AI search with AirPulse.

From audit finding to merged fix.

01

Connect

Checks the AirPulse MCP connection and matches the repository to one of your brands. No connection, no review.

02

Pull the audit

Reads your latest GEO audit, its findings and their tasks, and offers a fresh audit when the last one is old.

03

Confirm in code

Maps each finding to the file that renders the page and checks the HTML a crawler receives. Each finding gets a verdict.

04

Propose fixes

Shows each fix as a diff in the template that owns it, writes a report, and can mark tasks done in AirPulse.

124 checks, page by page.

Every check says where it lives in code, the rule, and how to test it on the page a crawler receives.

Access

Can engines fetch it?

robots.txt rules for search and AI crawlers, llms.txt, sitemaps and lastmod, pages that only render with JavaScript.

Meta data

Is the page described?

Titles, descriptions, canonical, robots tags, language and Open Graph, in the server HTML.

Schema

Is the markup right?

JSON-LD that parses, fits the page type and matches what the page shows.

Structure

Is it easy to parse?

One H1, heading order, question headings, real tables and lists.

Content

Can it be quoted?

Answer first, sections that stand on their own, specific facts with sources, prices as text.

Trust

Can it be believed?

Named authors, honest dates, about and contact pages, sources that resolve.

Connect the AirPulse MCP.

Add the server URL to your agent and sign in with your AirPulse login. Scripts and CI use an access token from the MCP page in the app; keep it out of the repository. The AirPulse MCP page shows what the assistant can read, and the release note has the steps for each assistant.

Claude Code
claude mcp add --transport http airpulse https://prod-api.airpulse.ai/mcp/
Codex, with an access token
export AIRPULSE_MCP_TOKEN=<token>
codex mcp add airpulse --url https://prod-api.airpulse.ai/mcp/ --bearer-token-env-var AIRPULSE_MCP_TOKEN

What it will not do.

Edit files without asking, or invent prices, reviews, authors or dates.
Change which crawlers your site allows. That stays your decision.
Send your code anywhere. It calls your AirPulse MCP and reads your own public pages.
Run without AirPulse. If the MCP is not connected, it stops and says how to connect.

Source, changelog and the full list of checks: github.com/airmeet/airpulse-skills.

Questions teams ask first

A folder of instructions and scripts that a coding agent such as Claude Code, Codex or Cursor loads when a task needs it. airpulse-site-review teaches the agent how to review a website's code for AI search, using your AirPulse data.

Bring your audit into the codebase.

AirPulse customers install the skill and connect the MCP. New to AirPulse? Book a free strategy call and we run your buyers' questions live.