Ever lets your AI coding agent drive a real Chrome browser — navigate, read the page, click, type, extract data, screenshot, and run autonomous web tasks. It works with Claude Code, Codex, Cursor, Gemini CLI, and ~30 other agents.
This repo is the install hub: a plugin (skill + MCP server) for Claude Code & Codex, a skill for every other agent, and the @everbrowser/cli package that powers them all.
Ever automates your Chrome, so two one-time steps are needed no matter how you install:
- Install the Chrome extension → Ever on the Chrome Web Store → Add to Chrome
- Sign in → click the Ever icon in the toolbar → the side panel opens → sign in with Google
If a command later says the browser is "not connected," it almost always means the extension isn't installed or you're not signed in — do the two steps above and retry.
One step wires up both the skill and the native browser_* tools (via MCP):
Claude Code
/plugin marketplace add namuh-eng/everbrowser
/plugin install ever-browser@everbrowser
Codex
codex plugin marketplace add namuh-eng/everbrowserThen enable ever-browser in the Codex plugin directory and restart.
npx skills add namuh-eng/everbrowserInstalls to the universal .agents/skills/ directory (covers Cursor, Codex, Gemini CLI, GitHub Copilot, Amp, Cline, OpenCode, Warp, and 30+ more), with Claude Code, Augment, and Continue available during setup.
npm install -g @everbrowser/cli # gives you `ever` and `ever mcp`
ever --versionPoint any MCP client at ever mcp (stdio). Full command reference: @everbrowser/cli on npm.
ever start --url https://example.com
ever snapshot
ever stopIf all three succeed, you're set. (If start reports "not connected," revisit the two prerequisite steps.)
your AI agent ──▶ Ever (CLI commands or MCP browser_* tools)
└──▶ everd (local daemon) ──▶ Chrome extension (CDP) ──▶ your browser
- Skill = teaches the agent to run
everCLI commands in the terminal. - MCP = gives the agent native
browser_*tools it calls directly (ever mcp). - Plugin = ships both in one install — best for Claude Code & Codex.
All paths drive the same daemon → extension → browser, and all need the prerequisites above.
ever start --url <url>— open a sessionever snapshot— capture the DOM with[id]annotations on interactive elementsever click <id>/ever input <id> "text"— act on elements by id- Re-snapshot after the page changes — old ids become stale
ever stop— end the session
For scripted automation and scraping, use ever exec (injects page + browser JS globals):
ever exec "await page.goto('https://example.com'); await browser.snapshot()"
ever exec --file ./scrape.jsRun ever --help for the full command set, or see the npm README.
| Skill | Description |
|---|---|
| ever-browser | Browser control commands, the exec scripting API, video recording, and scraping recipes |
A skill is just a folder with a SKILL.md and optional recipe scripts.
- Get it working interactively first with
evercommands andever exec. - Save reusable scripts under
skills/<name>/recipes/(see the x-feed-scraper recipe). - Write
SKILL.mdwith frontmatter (name,description) and clear sections: Quick Start, Core Workflow, Recipes, Error Recovery. - Open a PR to namuh-eng/everbrowser, or share from your own repo via
npx skills add <you>/<repo>.
If Ever is useful, a ⭐ on the repo helps others find it.