Skip to main content
PinchTab

​
What is PinchTab?

PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. It’s a 12MB Go binary that provides token-efficient browser automation via HTTP API or CLI.

Token-Efficient

800 tokens/page with text extraction - 5-13x cheaper than screenshots

Multi-Instance

Run multiple parallel Chrome processes with isolated profiles

Self-Contained

12MB binary, no external dependencies required

Accessibility-First

Stable element refs instead of fragile coordinates

​
Key Features

Control via command-line or HTTP API. Use curl, Python, Node.js, or any HTTP client.
# CLI
pinchtab instance launch

# HTTP API
curl -X POST http://localhost:9867/instances/launch
Run without a window (headless) or with visible Chrome (headed) for debugging.
# Headless (default, faster)
pinchtab instance launch --mode headless

# Headed (visible window)
pinchtab instance launch --mode headed
Browser profiles preserve cookies, sessions, and auth across instance restarts. Log in once, stay logged in.
# Create persistent profile
pinchtab profile create work

# Launch instance with profile
curl -X POST http://localhost:9867/instances/start \
  -d '{"profileId":"prof_278be873"}'
Run multiple instances in parallel with complete isolation - no shared state, no cookie leakage.
# Create 3 independent instances
for i in 1 2 3; do
  pinchtab instance launch --mode headless
done

# Each instance gets auto-allocated port: 9868, 9869, 9870

​
Why PinchTab?

For AI Agents

Designed specifically for AI-driven browser automation with token-efficient snapshots and stable element references.

For Developers

Simple HTTP API, CLI tools, and comprehensive documentation. Start automating in 5 minutes.

For Scale

Multi-instance architecture with auto-allocated ports (9868-9968). Run hundreds of parallel browsers.

For Security

Profile isolation, stealth mode, and secure token authentication for remote access.

​
How It Works

PinchTab consists of four core entities working together: 
PinchTab Orchestrator (HTTP server on port 9867)
  │
  ├── Instance 1 (inst_0a89a5bb, port 9868, temp profile)
  │     ├── Tab 1 (tab_xyz123, https://example.com)
  │     ├── Tab 2 (tab_xyz124, https://google.com)
  │     └── Tab 3 (tab_xyz125, https://github.com)
  │
  ├── Instance 2 (inst_1b9a5dcc, port 9869, profile: work)
  │     ├── Tab 1 (tab_abc001, internal dashboard, logged in)
  │     └── Tab 2 (tab_abc002, internal docs)
  │
  └── Instance 3 (inst_2c8a5eef, port 9870, profile: personal)
        ├── Tab 1 (tab_def001, gmail.com)
        └── Tab 2 (tab_def002, bank.com)
The orchestrator manages all instances and routes requests. Each instance is a separate Chrome process with optional profile for persistent state. Each tab is a webpage you navigate and interact with.

​
Quick Example

Here’s a complete browser automation workflow in just a few commands:
1

Start the orchestrator

pinchtab
Expected output:
🦀 Pinchtab Dashboard port=9867
dashboard ready url=http://localhost:9867/dashboard
2

Create a Chrome instance

# Create headless instance
INST=$(pinchtab instance launch --mode headless | jq -r '.id')
echo "Instance: $INST"

# Wait for Chrome to initialize
sleep 2
3

Navigate to a website

# Create tab and navigate
TAB_ID=$(curl -s -X POST http://localhost:9867/instances/$INST/tabs/open \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com"}' | jq -r '.id')
4

Get page content

# Get page structure with interactive elements
curl http://localhost:9867/tabs/$TAB_ID/snapshot | jq '.nodes | map({ref, role, name})'

# Extract text content
curl http://localhost:9867/tabs/$TAB_ID/text | jq '.text'
5

Interact with the page

# Click a button (ref from snapshot)
curl -X POST http://localhost:9867/tabs/$TAB_ID/action \
  -H "Content-Type: application/json" \
  -d '{"kind":"click","ref":"e5"}'

# Fill an input field
curl -X POST http://localhost:9867/tabs/$TAB_ID/action \
  -H "Content-Type: application/json" \
  -d '{"kind":"fill","ref":"e3","text":"[email protected]"}'
Check out the Quick Start guide to get PinchTab running in 5 minutes, or explore Core Concepts to understand the mental model.

​
Use Cases

AI Agent Automation

Build AI agents that can browse, click, fill forms, and extract data from any website.

Web Scraping

Extract text efficiently (~800 tokens/page) instead of expensive screenshots (10,000+ tokens).

End-to-End Testing

Automate browser testing with stable element references and multi-instance parallelization.

Multi-Account Management

Manage multiple user sessions with isolated profiles and persistent authentication.

​
Comparison

FeaturePinchTabPuppeteerPlaywrightSelenium
Token efficiency✅ 800 tokens/page❌ 10k+ tokens❌ 10k+ tokens❌ 10k+ tokens
HTTP API✅ Native❌ Library only❌ Library only❌ Library only
Multi-instance✅ Auto-managed⚠️ Manual⚠️ Manual⚠️ Manual
Persistent profiles✅ Built-in⚠️ Manual⚠️ Manual⚠️ Manual
Binary size✅ 12MB❌ Node deps❌ Node deps❌ Java deps
Stealth mode✅ Built-in⚠️ Plugins✅ Built-in❌ Not available
Element stability✅ Accessibility refs❌ Selectors❌ Selectors❌ XPath
PinchTab is specifically designed for AI agents and token efficiency, making it ideal for LLM-driven automation workflows.

​
Next Steps

Quick Start

Get PinchTab running in 5 minutes with installation and first commands

Core Concepts

Understand orchestrator, instances, profiles, and tabs

API Reference

Complete HTTP API documentation with all endpoints

Examples

Real-world examples and code snippets

​
Community & Support

GitHub

Star the repo, report issues, contribute code

Discussions

Ask questions, share use cases, get help

Build docs developers (and LLMs) love

Get started for freeTalk to us