# Best open-source AI repos for browser automation with AI agents (2026)

Source: https://dedupedai.com/best-ai-repos-for/browser-automation/ (Deduped, updated 2026-10-27). Markdown twin of the page; JSON: https://dedupedai.com/best-ai-repos-for/browser-automation/index.json

## What is the best open-source repo for AI browser automation?

The best open-source AI repos for browser automation with AI agents right now are Playwright MCP, Chrome DevTools MCP and Browser Use. To let a coding agent drive a real browser, use Playwright MCP or Chrome DevTools MCP; for a scriptable agent that browses by itself, use browser-use; for a fast CLI built for agents, use agent-browser.

## Comparison

| Repo | What it does | Best for | Stars | Language |
|---|---|---|---|---|
| [Playwright MCP](https://dedupedai.com/use-cases/microsoft-playwright-mcp/) (microsoft/playwright-mcp) | Playwright MCP is Microsoft's MCP server for browser automation with Playwright. | Developers and QA engineers who want an agent to click through a web app, test flows or collect data in a real browser. | 38,601 | TypeScript |
| [Chrome DevTools MCP](https://dedupedai.com/use-cases/chromedevtools-chrome-devtools-mcp/) (ChromeDevTools/chrome-devtools-mcp) | Chrome DevTools for agents (chrome-devtools-mcp) lets a coding agent such as Claude, Cursor, Copilot or Antigravity control and inspect a live Chrome browser, as an MCP server or as a CLI. | Web developers who want their coding agent to see what the page actually does in Chrome instead of guessing from source. | 54,099 | TypeScript |
| [Browser Use](https://dedupedai.com/use-cases/browser-use-browser-use/) (browser-use/browser-use) | browser-use is an open-source Python library for AI browser agents that navigate websites with any LLM, plus an optional cloud with stealth browsers and a hosted agent API. | Developers automating web tasks such as form filling, lookups and bookings with AI agents. | n/a | Python |
| [Agent Browser](https://dedupedai.com/use-cases/vercel-labs-agent-browser/) (vercel-labs/agent-browser) | agent-browser is a fast native Rust CLI from Vercel Labs for browser automation by AI agents. | Developers who want their coding agent to use a browser through short shell commands instead of an MCP server. | 44,552 | Rust |
| [Cua](https://dedupedai.com/use-cases/trycua-cua/) (trycua/cua) | Cua gives AI agents computers they can use. | Developers building or running computer-use agents who need safe desktops, sandboxes and app automation instead of letting agents loose on their main machine. | 31,034 | Rust |
| [Jev Ultrafast](https://dedupedai.com/use-cases/browser-use-jev-ultrafast/) (browser-use/jev-ultrafast) | Jev Ultrafast is a browser agent from Browser Use with a dynamic, indexed action space. | Developers building fast web agents who want fewer model calls, no screenshots in the loop and a codebase small enough to read. | 23,686 | Python |
| [Ego Lite](https://dedupedai.com/use-cases/citrolabs-ego-lite/) (citrolabs/ego-lite) | ego lite is a browser built for you and your AI agents to share. | People who use coding agents for web tasks (research, lead enrichment, form filling) and want them to use their real logins without taking over the browser. | 19,746 | JavaScript |

## When to pick which

### Playwright MCP (microsoft/playwright-mcp)

Playwright MCP is Microsoft's MCP server for browser automation with Playwright. It lets an LLM work with web pages through structured accessibility snapshots instead of screenshots, so no vision model is needed. It installs with one npx command in Claude Code, Codex, Cursor, VS Code, Amp and other clients, and supports Chrome, Firefox, WebKit and Edge. The README suggests the Playwright CLI with skills for token-heavy coding agent work and MCP for exploratory, stateful agent loops.

Best for: Developers and QA engineers who want an agent to click through a web app, test flows or collect data in a real browser.

Needs: Node.js; An MCP client such as Claude Code, Codex, Cursor or VS Code.

Start here: [Give Claude Code or Codex a real browser in one command](https://dedupedai.com/use-cases/microsoft-playwright-mcp/add-a-browser-to-claude-code-or-codex/) (Beginner, 5 minutes)

1. Claude Code: claude mcp add playwright npx @playwright/mcp@latest
2. Codex: codex mcp add playwright npx "@playwright/mcp@latest"
3. Other clients: add an mcpServers entry named playwright that runs npx with @playwright/mcp@latest
4. Ask the agent to open your app and walk through a flow

All 3 use cases: https://dedupedai.com/use-cases/microsoft-playwright-mcp/ | GitHub: https://github.com/microsoft/playwright-mcp

### Chrome DevTools MCP (ChromeDevTools/chrome-devtools-mcp)

Chrome DevTools for agents (chrome-devtools-mcp) lets a coding agent such as Claude, Cursor, Copilot or Antigravity control and inspect a live Chrome browser, as an MCP server or as a CLI. The agent can record performance traces and get insights, read network requests and console messages with source-mapped stack traces, take screenshots, and automate pages with Puppeteer. The README warns that it exposes the browser's content to the MCP client, so avoid sensitive sites in that browser.

Best for: Web developers who want their coding agent to see what the page actually does in Chrome instead of guessing from source.

Needs: Node.js and npm; Chrome; An MCP client or a coding agent that can run shell commands.

Start here: [Let your coding agent debug a page in a real Chrome](https://dedupedai.com/use-cases/chromedevtools-chrome-devtools-mcp/let-your-agent-debug-a-page-in-real-chrome/) (Beginner, 10 minutes)

1. Add an mcpServers entry named chrome-devtools that runs npx with chrome-devtools-mcp@latest
2. Restart your MCP client so the tools load
3. Ask the agent to open your local app and report console messages and failed network requests
4. Have it fix the cause and take a screenshot to confirm

All 3 use cases: https://dedupedai.com/use-cases/chromedevtools-chrome-devtools-mcp/ | GitHub: https://github.com/ChromeDevTools/chrome-devtools-mcp

### Browser Use (browser-use/browser-use)

browser-use is an open-source Python library for AI browser agents that navigate websites with any LLM, plus an optional cloud with stealth browsers and a hosted agent API.

Best for: Developers automating web tasks such as form filling, lookups and bookings with AI agents.

Needs: Python 3.12 with uv; An LLM API key such as OpenAI, or a Browser Use key.

Start here: [Automate a web task with a Python browser agent](https://dedupedai.com/use-cases/browser-use-browser-use/automate-web-task-with-python-agent/) (Intermediate, 20 minutes)

1. With uv installed, start a project with uv init --python 3.12 if needed
2. Add the library: uv add browser-use
3. Put OPENAI_API_KEY in .env (or BROWSER_USE_API_KEY to use the BU2 model)
4. In agent.py create Agent(task="...", llm=ChatOpenAI(...)), call await agent.run() and print history.final_result()

All 3 use cases: https://dedupedai.com/use-cases/browser-use-browser-use/ | GitHub: https://github.com/browser-use/browser-use

### Agent Browser (vercel-labs/agent-browser)

agent-browser is a fast native Rust CLI from Vercel Labs for browser automation by AI agents. An agent opens a page, takes an accessibility snapshot with element refs, then clicks, fills and reads by ref, and takes screenshots, including annotated ones and screenshots skipped when nothing changed to save tokens. It downloads Chrome for Testing on first run or uses an existing Chrome, Brave, Playwright or Puppeteer install.

Best for: Developers who want their coding agent to use a browser through short shell commands instead of an MCP server.

Needs: npm, Homebrew or Cargo; Chrome (agent-browser install downloads Chrome for Testing).

Start here: [Let your agent browse with short CLI commands](https://dedupedai.com/use-cases/vercel-labs-agent-browser/let-your-agent-browse-with-short-cli-commands/) (Beginner, 10 minutes)

1. Install: npm install -g agent-browser, then agent-browser install to get Chrome
2. Open a page: agent-browser open example.com
3. Get the accessibility tree with refs: agent-browser snapshot
4. Act by ref: agent-browser click @e2, agent-browser fill @e3 "test@example.com", agent-browser get text @e1

All 3 use cases: https://dedupedai.com/use-cases/vercel-labs-agent-browser/ | GitHub: https://github.com/vercel-labs/agent-browser

### Cua (trycua/cua)

Cua gives AI agents computers they can use. The repository holds Cua Spaces (a desktop app that gives agents full macOS or Linux desktops on your Mac, other machines you own or your own cloud), Cua Driver for operating native apps and browsers, Lume for local macOS and Linux VMs on Apple Silicon, a Cua SDK and CLI for sandboxes, CUA-S1 decision models and Cua Bench for evaluating computer-use agents.

Best for: Developers building or running computer-use agents who need safe desktops, sandboxes and app automation instead of letting agents loose on their main machine.

Needs: macOS 26 or later for Cua Spaces; Apple Silicon for Lume VMs; macOS, Linux or Windows for the cua CLI and Cua Driver; Spaces is free for individuals; Pro and Teams are paid.

Start here: [Spin up a disposable Linux sandbox for an agent](https://dedupedai.com/use-cases/trycua-cua/spin-up-a-sandbox-for-an-agent/) (Beginner, 15 minutes)

1. Install with curl -fsSL https://cua.ai/install.sh | sh (Windows uses the PowerShell installer)
2. Create a sandbox: cua sb create ubuntu --name dev
3. Run a command inside: cua sb exec dev uname -a
4. Take a screenshot with cua sb screenshot dev

All 3 use cases: https://dedupedai.com/use-cases/trycua-cua/ | GitHub: https://github.com/trycua/cua

### Jev Ultrafast (browser-use/jev-ultrafast)

Jev Ultrafast is a browser agent from Browser Use with a dynamic, indexed action space. Each page becomes a numbered element table, TypeSafe's Jev picks one operation (CLICK, TYPE_TEXT, SELECT, SCROLL, WAIT, DONE or BLOCKED) and a target in one network round trip, and a small LLM writes text only when the operation is TYPE_TEXT. The README shows a Zurich to London Google Flights search in about 7 seconds.

Best for: Developers building fast web agents who want fewer model calls, no screenshots in the loop and a codebase small enough to read.

Needs: Python with uv; Chrome with remote debugging allowed (connected through Browser Harness, installed by uv sync); TYPESAFE_API_KEY and TEXT_MODEL_API_KEY (an OpenRouter key in the example config); Live examples make paid API calls.

Start here: [Run a 7 second browser agent and watch every decision](https://dedupedai.com/use-cases/browser-use-jev-ultrafast/watch-a-fast-browser-agent-in-the-inspector/) (Intermediate, 20 minutes)

1. Run git clone https://github.com/browser-use/jev-ultrafast.git and cd jev-ultrafast
2. Run uv sync then cp .env.example .env and add TYPESAFE_API_KEY and TEXT_MODEL_API_KEY
3. Start the inspector with uv run jev and open http://127.0.0.1:8766
4. Click Start demo, then Run automatically; use Choose next to pause before each action

All 3 use cases: https://dedupedai.com/use-cases/browser-use-jev-ultrafast/ | GitHub: https://github.com/browser-use/jev-ultrafast

### Ego Lite (citrolabs/ego-lite)

ego lite is a browser built for you and your AI agents to share. Each agent runs browser tasks in its own isolated Space inside the same browser while you keep browsing, it exposes page actions to the agent as JavaScript functions for faster runs on fewer tokens, and the ego-browser skill lets Claude Code, Codex and other agents use your real logins and tabs.

Best for: People who use coding agents for web tasks (research, lead enrichment, form filling) and want them to use their real logins without taking over the browser.

Needs: macOS or Windows (Linux is on the roadmap); An agent CLI that supports skills, such as Claude Code or Codex; Optional: migrate Chrome data on first launch to inherit logins.

Start here: [Let your coding agent do web tasks with your real logins](https://dedupedai.com/use-cases/citrolabs-ego-lite/let-your-agent-browse-with-your-logins/) (Beginner, 15 minutes)

1. Run npx skills add citrolabs/ego-lite to add the ego-browser skill, or download the app
2. On first launch, choose whether to migrate your Chrome data so the agent inherits logins
3. In your agent CLI type /ego-browser followed by a plain language task
4. Watch the agent open the page in its own Space, read a Snapshot and act

All 3 use cases: https://dedupedai.com/use-cases/citrolabs-ego-lite/ | GitHub: https://github.com/citrolabs/ego-lite

## Questions

### What is the best open-source repo for AI browser automation?

The best open-source AI repos for browser automation with AI agents right now are Playwright MCP, Chrome DevTools MCP and Browser Use. To let a coding agent drive a real browser, use Playwright MCP or Chrome DevTools MCP; for a scriptable agent that browses by itself, use browser-use; for a fast CLI built for agents, use agent-browser.

### Which repo for browser automation with AI agents is easiest to start with?

Playwright MCP: its starter use case "Give Claude Code or Codex a real browser in one command" is rated Beginner and takes about 5 minutes.

### How do I get started with Playwright MCP?

Claude Code: claude mcp add playwright npx @playwright/mcp@latest; Codex: codex mcp add playwright npx "@playwright/mcp@latest"; Other clients: add an mcpServers entry named playwright that runs npx with @playwright/mcp@latest. Full steps and a copy-paste agent prompt: https://dedupedai.com/use-cases/microsoft-playwright-mcp/add-a-browser-to-claude-code-or-codex/

### Are these browser automation with AI agents repos free?

They are open-source projects on GitHub (check each license; some, like fair-code projects, limit commercial use). Some also sell a hosted version or need a paid model API key; the requirements for each repo are listed on this page.

### How is this list of browser automation with AI agents repos chosen?

Deduped tracks the hottest AI repos on GitHub and writes step-by-step use cases from each README. For this page we picked the repos that do this job, ordered by fit and then GitHub stars. Repos we only inferred from a short README, or that we do not index, are left out. The page is rebuilt with the site; stars are as of the last build in 2026.

Third-party repos we don't own or vouch for. AI-written use cases can be wrong: review everything and run at your own risk.