Lighthouse Agentic Browsing · WebMCP · llms.txt · Free
Agentic Browsing Skill
A Claude Code skill that audits and fixes any website for Google Lighthouse's Agentic Browsing category — so AI agents and answer engines can read, navigate and operate it, not just rank it. Covers crawlable HTML, the agent-centric accessibility tree, layout stability, llms.txt, and the WebMCP build-out. The exact playbook that took puniverse.net to a full agentic-browsing pass.
Watch the 60-second intro
Who's it for?
For founders and developers shipping React, Next, Vite or Astro sites who want to be discoverable and usable in the agentic web — ChatGPT, Perplexity, Claude, Copilot and operator agents — not just ranked in search.
How it works
Run the included zero-dependency Node audit on any URL to flag the scored signals in seconds.
Fix the crawlable-HTML foundation so the page is server-rendered and the accessibility tree exists before JavaScript runs.
Tidy the agent-centric accessibility tree: names, labels, heading order, roles and visibility.
Stabilise layout (CLS) so agents act on elements that do not move between identifying and clicking them.
Add a well-formed llms.txt, then expose forms, search and checkout as WebMCP tools agents can drive.
How to set up
- Download the skill from Gumroad (free) and unzip it.
- Copy the agentic-browsing folder into your Claude Code skills directory (~/.claude/skills/).
- Start a new Claude Code session — the skill triggers when you ask to make a site agent-ready.
- Run the audit script against your site and apply the fixes it surfaces.
- Run a real Lighthouse Agentic Browsing report in Chrome 150+ to confirm the score.