@ulpi/browse
v1.1.1 — React DevTools integration

Fast headless browser for AI agents

The CLI that lets AI agents browse the web. 2.4x fewer tokens than alternatives. ~100ms per command.

$npx @ulpi/browse
terminal
$

2.4x

fewer tokens

vs browser-use and @playwright/mcp

~100ms

ms per command

After initial startup

128

commands

Navigation, interaction, extraction, and more

MIT

MIT License

Free and open source

Built for AI agents

Token efficient

Selective snapshot loading means agents only request accessibility trees when needed — not on every action.

Blazing fast

Persistent daemon architecture. First command starts the server, subsequent commands execute in ~100ms.

@ref system

No CSS selectors needed. Snapshot assigns numbered refs (@e1, @e2) to elements. Click @e3, fill @e5.

Works everywhere

Claude Code, Cursor, Cline, Windsurf — any AI agent that can run bash commands.

Model Context Protocol

Native MCP integration

Run browse as an MCP server. Cursor, Claude Desktop, and Windsurf call browse tools directly — no shell commands, no wrappers. Same 128 commands, same ~100ms speed.

CursorClaude DesktopWindsurf
View MCP setup guide
.cursor/mcp.json
{
  "mcpServers": {
    "browse": {
      "command": "browse",
      "args": ["--mcp"]
    }
  }
}

Local or cloud. Your choice.

Switch browser engines on the fly or offload to cloud providers. Same commands work everywhere.

Local

Playwright (default)

Full-featured Chromium with auto-install. All 128 commands, most tested and stable. The default for local development.

Local

Rebrowser (stealth)

Patched Chromium that evades bot detection. Passes Cloudflare, DataDome, and more. Same API, drop-in replacement.

Local

LightPanda (lightweight)

Zig-based browser engine. 10x faster page loads, ~10MB memory. Ideal for static content extraction and simple scraping.

Local

Camoufox (anti-detection)

Firefox with C++ fingerprint spoofing. Bypasses Cloudflare Turnstile, bot detection, and fingerprinting. Named profiles for per-scenario configs.

Cloud

Browserbase

Managed cloud Chromium for CI/CD and serverless. Spin up sessions on demand — no local browser needed.

Cloud

Browserless

Connect to Browserless for scalable headless browsing. API keys encrypted at rest, never visible to agents.

Explore runtimes & providers
Terminal
$ browse react-tree
# App > Layout > Dashboard
#   <UserProfile name="Alice">
#   <DataTable rows=142 loading=false>

$ browse react-props @e12
# { rows: 142, onSort: fn, filter: "active" }

$ browse react-state @e12
# { sortCol: "date", page: 3 }
React DevTools

Your agent can debug React — on the fly

Inspect component trees, read props and state, find Suspense boundaries, and measure render timing. Your AI agent can now diagnose UI bugs without you lifting a finger.

  • Inspect the full React component tree with one command
  • Read any component's props, state, and hooks in real time
  • Find Suspense boundaries and hydration mismatches automatically
  • Measure render timing and identify performance bottlenecks
View React DevTools guide

Works with your favorite AI agent

Compatible with 39 AI coding agents and IDEs

AdaL
Amp
Antigravity
Augment
Claude Code
Cline
CodeBuddy
Codex
Command Code
Continue
Cortex Code
Crush
Cursor
Droid
Gemini CLI
GitHub Copilot
Goose
iFlow CLI
Junie
Kilo Code
AdaL
Amp
Antigravity
Augment
Claude Code
Cline
CodeBuddy
Codex
Command Code
Continue
Cortex Code
Crush
Cursor
Droid
Gemini CLI
GitHub Copilot
Goose
iFlow CLI
Junie
Kilo Code
Kimi Code CLI
Kiro CLI
Kode
MCPJam
Mistral Vibe
Mux
Neovate
OpenClaw
OpenCode
OpenHands
Pi
Pochi
Qoder
Qwen Code
Roo Code
Trae
Warp
Windsurf
Zencoder
Kimi Code CLI
Kiro CLI
Kode
MCPJam
Mistral Vibe
Mux
Neovate
OpenClaw
OpenCode
OpenHands
Pi
Pochi
Qoder
Qwen Code
Roo Code
Trae
Warp
Windsurf
Zencoder
v1.4.0 — Performance Audit

Full performance audit in one command

Core Web Vitals, LCP critical path, layout shift attribution, JS/CSS coverage, 108 framework detections, 55 SaaS platforms. Built for AI agents.

108
Framework signatures
55
SaaS platforms
<200ms
Detection speed
$npm install -g @ulpi/browse
$ browse perf-audit shopify-store.com

── Core Web Vitals ──────────────────────
LCP2.9s  △ needs improvement
CLS0.02  ✓ good
TBT820ms  ✗ poor
FCP1.3s  ✓ good
TTFB180ms  ✓ good
INP310ms  △ needs improvement

── Tech Stack ────────────────────────────
PlatformShopify (Dawn 12.0)
Apps23 apps · 1.4MB JS
CDNCloudflare (h2)

── Recommendations ──────────────────────
→ Remove 8 unused Shopify apps — 620ms TBT
→ Preload LCP hero image — saves ~500ms
→ 4 fonts missing font-display: swap

Start browsing in seconds

One command to install. Zero configuration.