Phyll

phyll

UX review for apps built with AI, inside the agent you already use.

Phyll plugs into Codex, Claude Code or any agent that speaks MCP. Your agent opens your app the way a first-time user would, walks the core jobs in a real browser, and writes a report of what gets in the way: forms that ask for more than the job needs, sign-up walls, dead buttons, text too faint to read. The fixes keep your design.

The AI work runs in your agent, on your own plan. This package adds the browser, the page probe, the source scanner and the link to Phyll's engine, which sends the review method, checks the report and keeps it with a link.

Start

npx phyll signup [email protected]npx phyll setup codex        # or claude, cursor, windsurf, gemini

Then ask your agent: "review my app at http://localhost:3000".

setup registers the connector in Codex's config.toml, with claude mcp add, or in the mcpServers of Cursor's ~/.cursor/mcp.json, Windsurf's ~/.codeium/windsurf/mcp_config.json or Gemini CLI's ~/.gemini/settings.json, keeping everything else in the file. It also installs the Chromium build the connector uses. For any other agent that takes MCP servers, npx phyll setup other prints the entry to paste. You need Node 20 or newer.

Your reports, keys and plan are also on the site: npx phyll account opens your account there, already signed in. If you lose the key, sign in at agentphyll.com/login with your email and create a new one.

Commands

The tools your agent gets

start_review scans the project, creates the report folder and returns the method. capture saves a screenshot and the page probe of each route at laptop and phone size. open, snapshot, click, fill, select, check, press, back, screenshot, probe and resize drive the browser, which stays on your app. guide returns a part of the method, finish_review sends the report to the engine and writes report.md, and scan, account and upgrade do what their names say.

What leaves your computer

Your source code, your screenshots and your agent's conversation stay on your computer. When a review starts, the connector sends the app's address, the project name and a summary of the scan: which tells it found, how many times, and the paths of routes and forms. When it ends, it sends the report your agent wrote, which the engine keeps so the link works.

More

The scanner, the plugin and four example apps are in the repository. MIT license.