PinchTab — Browser Automation Bridge for AI Agents
PinchTab is a 9MB Go binary that gives AI agents direct control over Chrome. Here's how I installed it, tested it against my own site, and plan to use it with n8n workflows.
PinchTab — Browser Automation Bridge for AI Agents
I came across PinchTab today — a 9MB Go binary that gives AI agents direct control over a Chrome browser via HTTP. Created on February 15, 2026 and already at 2,800+ stars. Here’s what it does and how I got it running in minutes.
What you’ll have at the end:
- ✅ PinchTab running locally on your Mac
- ✅ AI agents controlling Chrome via HTTP API
- ✅ 5-13x cheaper page extraction than screenshots
- ✅ Ready to integrate with n8n workflows
Time required: ~5 minutes Cost: Free (MIT licence) Language: Go (single binary, no dependencies)
What is PinchTab?
PinchTab is a standalone HTTP server that lets AI agents control a Chrome browser programmatically. Instead of taking screenshots (expensive in tokens), it extracts the accessibility tree — clean structured text at around 800 tokens per page.
1
2
3
4
5
6
7
AI Agent / n8n Workflow
↓
HTTP API (:9867)
↓
PinchTab
↓
Chrome Browser
It supports both headless and headed Chrome, persistent sessions, stealth injection to avoid bot detection, and multi-instance orchestration for parallel tasks.
Installation — macOS Apple Silicon (M3/M4)
Step 1 — Download the binary
1
2
curl -L -o pinchtab https://github.com/pinchtab/pinchtab/releases/download/v0.7.6/pinchtab-darwin-arm64
chmod +x pinchtab
Step 2 — Add to PATH
1
sudo cp pinchtab /usr/local/bin/pinchtab
Step 3 — Verify
1
2
pinchtab --version
# pinchtab 0.7.6
That’s it. Single binary, no dependencies, no Docker required.
Quick Test
Terminal 1 — Start the server:
1
2
pinchtab
# Starts on http://127.0.0.1:9867
Terminal 2 — Test it:
1
2
# Navigate to a URL
pinchtab nav https://davidtkeane.com
Output:
1
2
3
4
{
"title": "Ranger Products - Transform Disabilities into Superpowers | David T. Keane",
"url": "https://davidtkeane.com/"
}
Works immediately. No config, no API keys, no setup.
Full Command Reference
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
# Navigation
pinchtab nav <url> # Navigate to URL
# Page inspection
pinchtab snap -i -c # Interactive elements only (compact)
pinchtab snap -d # Only changes since last snapshot
pinchtab text # Extract all readable text
pinchtab text --raw # Raw text extraction
# Interaction
pinchtab click <ref> # Click element by reference
pinchtab type <ref> <text> # Type into element
pinchtab press <key> # Press key (Enter, Tab, Escape)
pinchtab fill <ref> <text> # Fill input directly
pinchtab hover <ref> # Hover element
pinchtab scroll <ref|pixels> # Scroll to element or by pixels
pinchtab select <ref> <value> # Select dropdown option
pinchtab focus <ref> # Focus element
# Output
pinchtab ss -o screenshot.png # Screenshot
pinchtab pdf -o page.pdf # Export as PDF
pinchtab eval "document.title" # Run JavaScript
# Management
pinchtab tabs # List tabs
pinchtab tabs new <url> # Open new tab
pinchtab tabs close <id> # Close tab
pinchtab health # Check server status
pinchtab dashboard # Profile orchestrator UI
Token Efficiency — Why This Matters
| Method | Tokens per page | Cost ratio |
|---|---|---|
| Screenshot (GPT-4o) | ~4,000–10,000 | Expensive |
| PinchTab accessibility tree | ~800 | 5-13x cheaper |
For AI agents that browse many pages, this is a significant saving. Instead of sending a PNG to a vision model, you send clean structured text.
Environment Variables
1
2
3
4
PINCHTAB_URL=http://127.0.0.1:9867 # Server URL (for CLI)
BRIDGE_TOKEN=your-secret-token # Auth token (optional)
BRIDGE_PORT=9867 # Change port
BRIDGE_HEADLESS=true # Run headless (default)
Integration with n8n
Since PinchTab exposes an HTTP API on port 9867, you can control it directly from n8n using HTTP Request nodes:
1
2
3
n8n HTTP Request node
→ POST http://127.0.0.1:9867/navigate
→ Body: { "url": "https://example.com" }
Combine with the AI Agent node in n8n for fully autonomous web browsing workflows — navigate, extract, interact, repeat.
Use Cases
- Web scraping — extract structured data from any site
- AI agent browsing — let your AI agents navigate and interact with websites
- Automated testing — click through flows and verify pages
- Price monitoring — check token/crypto prices on DEX sites
- Form automation — fill and submit forms programmatically
- PDF generation — export any web page as a clean PDF
Linux / VPS Installation
1
2
3
4
5
6
7
# Download linux-amd64 binary
curl -L -o pinchtab https://github.com/pinchtab/pinchtab/releases/download/v0.7.6/pinchtab-linux-amd64
chmod +x pinchtab
sudo mv pinchtab /usr/local/bin/
# Run headless (required on VPS — no display)
BRIDGE_HEADLESS=true pinchtab
Works on Ubuntu 24.04. Needs Chrome or Chromium installed:
1
sudo apt install -y chromium-browser
Resources
Support This Content
If this guide saved you time, consider buying me a coffee!
Written 2026-03-02 — tested on M3 Pro Mac with PinchTab v0.7.6. Single binary install, working in under 5 minutes.
Rangers lead the way! 🎖️