Browser automation
in 50KB
Zero dependencies. 70+ commands. Smart interactions — no CSS selectors, no LLM needed.
Built-in MCP server for AI agents. 500x fewer tokens than screenshots.
Works with Claude Code · Cursor · Windsurf · Any MCP client
Everything you need for
browser automation
70+ commands covering navigation, interaction, debugging, network control, and AI agent integration.
Navigate & Interact
go, click, fill, type, submit, hover, drag, scroll, and 40+ more commands. Full browser control from your terminal.
AI Agent Integration
Built-in MCP server with a11y-snapshot and click-ref. Give your AI agents browser superpowers in one line.
Visual Feedback
Green glow overlay, cursor visualization, click ripples, and keystroke display. See exactly what's happening.
Multi-Project Isolation
Each project gets its own browser instance with isolated cookies, storage, and sessions. No cross-contamination.
Annotated Screenshots
@N badges on interactive elements for AI vision. Your AI agent can see and reference every clickable element.
Batch & Auto-Wait
Pipe JSON commands for complex workflows. Built-in smart waiting for elements, navigation, and network idle.
Smart Commands
Click, fill, and select by visible text — like Stagehand's act() but without LLM cost. Fuzzy matching across labels, placeholders, and aria attributes.
How cdpilot compares
Lightweight by design. Powerful by default.
| Tool | Install Size | Dependencies | Setup | CLI-First | MCP Server | AI-Ready |
|---|---|---|---|---|---|---|
| cdpilot | 50KB | 0 | Instant | |||
| Playwright | ~300MB | 200+ | Minutes | |||
| Puppeteer | ~400MB | 80+ | Minutes | |||
| Browser Use | ~150MB | 40+ | Minutes |
The real cost of browser automation
Same task: Search Google, click first result
Computer Use
250,000 tokens
~$0.75/task
- 8+ LLM calls
- ~30 seconds
- Screenshot-based pixel analysis
Playwright MCP
114,000 tokens
~$0.34/task
- 4+ LLM calls
- ~15 seconds
- DOM dump CSS selectors
cdpilot
500 tokens
~$0.0015/task
- 2 LLM calls
- ~3 seconds
- Structured a11y @ref navigation
500x fewer tokens. Same result.
Give your AI agent
browser superpowers
Add one block to your MCP config and your AI can browse, click, fill forms, take screenshots, and extract structured data.
- One-line MCP setup — no config files, no API keys
- a11y-snapshot gives your AI a semantic view of any page
- click-ref lets AI click elements by @N reference
- Isolated browser sessions — no access to your personal data
{ "mcpServers": { "cdpilot": { "command": "npx", "args": ["cdpilot", "mcp"] } }}Simple, transparent pricing
Start free, upgrade when you need more.
Free
Everything you need to get started with browser automation.
- Forever free — no limits, no signup
- 60+ CLI commands
- 10 built-in test assertions
- Built-in MCP server (Claude Code, Cursor)
- a11y-snapshot (500x fewer tokens)
- Token-efficient screenshots (element crop, JPEG)
- Multi-project browser isolation
- Visual feedback (green glow, cursor, ripples)
- Pre-flight wizard (auto-setup)
- Unlimited usage
- Community support
Pro
Advanced features for power users and teams.
- Everything in Free
- Stealth Mode (human-like mouse movement, typing randomization)
- Anti-fingerprint (proxy rotation, UA rotation, canvas noise)
- CAPTCHA solving (2Captcha/hCaptcha integration)
- Parallel orchestration (10+ concurrent sessions)
- Workflow recorder & replay
- Priority support
- Early access to new features
Cloud
Hosted browser sessions with API access for CI/CD.
- Everything in Pro
- Hosted browser sessions (no local install)
- MCP WebSocket endpoint (wss://api.cdpilot.ndr.ist/mcp)
- REST API access
- Usage dashboard & analytics
- Auto-scaling
- 99.9% uptime SLA