Key Takeaways
PinchTab gives AI agents direct, token-efficient control over real Chrome browsers through a 12MB self-hosted binary, combining advanced stealth injection, persistent profiles, multi-instance orchestration, and a live dashboard to slash costs and eliminate detection risks.
In the rapidly evolving world of AI agents, browser automation remains one of the most critical yet challenging components. Traditional tools built for testing often fall short when powering autonomous agents that must navigate dynamic websites, maintain logins, extract structured data, and evade bot detection—all while keeping token consumption low for large language models. PinchTab addresses these pain points head-on as a standalone HTTP server that bridges AI agents to a full Chrome browser instance.
Released as a high-performance Go binary, PinchTab weighs just 12 megabytes, requires no external dependencies, and exposes a clean HTTP API alongside a command-line interface. It supports both headless and headed modes, runs natively on ARM64 devices including Raspberry Pi, and emphasizes accessibility-tree mapping over fragile DOM selectors or expensive screenshots. The result is a tool optimized specifically for agentic workflows rather than human QA scripts.
This guide covers everything you need to get started with PinchTab, from installation to advanced multi-instance orchestration and integration with AI agents. Whether you are building autonomous web agents, scaling data extraction pipelines, or deploying self-hosted automation stacks, PinchTab provides the reliability, stealth, and observability modern AI systems demand.
The Problem PinchTab Solves in AI Browser Automation
Modern AI agents excel at reasoning and planning but struggle with direct web interaction. Sending full HTML or screenshots to large language models consumes thousands of tokens per page and introduces latency. Coordinate-based automation breaks with the slightest layout change. Bot detection services flag common automation libraries within seconds. Persistent sessions across restarts are difficult to maintain without custom infrastructure.
PinchTab solves these issues by acting as a thin, high-performance orchestrator. It launches real Chrome processes via the Chrome DevTools Protocol, extracts information through the browser’s accessibility tree (averaging only 800 tokens per page), and injects stealth patches at the protocol level. Profiles store cookies, local storage, and extensions persistently, while the built-in orchestrator manages multiple isolated Chrome instances simultaneously. A real-time dashboard provides live visibility into every running instance, tab, and resource.
Developers and AI engineers report 5-13× lower token usage compared with screenshot-based approaches, dramatically extending the practical length of agent sessions and reducing inference costs.
Core Features That Set PinchTab Apart
PinchTab’s design philosophy centers on simplicity, efficiency, and agent-friendliness. Several standout capabilities make it uniquely suited for production AI workflows.
Accessibility-First Control with Stable References
Instead of relying on brittle XPath or CSS selectors, PinchTab exposes the browser’s accessibility tree with deterministic element references such as “e5” or “e12”. Agents click, type, or read content using these stable refs, eliminating the need for repeated DOM parsing or visual analysis.
Token-Efficient Data Extraction
The default text endpoint returns clean, readability-mode content stripped of navigation and advertisements. Snapshot endpoints return structured JSON of the accessibility tree. Both approaches keep context windows small and costs low—ideal for repeated agent loops or long-running tasks.
Advanced Stealth Injection
PinchTab automatically patches common detection vectors: navigator.webdriver, navigator.plugins, WebGL fingerprinting, and automation flags. User-agent spoofing and protocol-level hiding ensure sessions pass Cloudflare, PerimeterX, and other major bot mitigation systems. Because the patches occur inside the real Chrome process, detection resistance exceeds that of most wrapper libraries.
Real-Time Dashboard and Orchestration
The HTTP server running on port 9867 (configurable) doubles as a real-time dashboard. Administrators and agents alike can monitor all active instances, profile usage, tab states, and performance metrics through a unified web interface or API calls. The orchestrator handles lifecycle management—launching, scaling, and terminating isolated Chrome processes—without manual port juggling.
Multi-Instance and Profile Isolation
Create as many parallel Chrome instances as hardware allows, each with its own profile directory. One instance can handle an Alice account while another manages Bob’s session, with zero cookie leakage. Profiles survive restarts, enabling true session persistence across agent runs.
Persistent Sessions and Multi-Tab Support
Log in once through a headed window or script; the profile retains authentication indefinitely. Each instance supports multiple tabs, allowing complex workflows such as opening a dashboard while processing detail pages in background tabs.
Framework-Agnostic API and CLI
Control everything via simple curl commands, the official CLI, or any programming language. Integration with existing agent frameworks requires only HTTP requests—no SDK lock-in.
Additional utilities include JavaScript evaluation as an escape hatch, JPEG screenshots with quality control, and full-page PDF export for archiving.
Installing PinchTab on Any Platform
PinchTab’s installation process takes under 30 seconds on most systems. Three official methods cover every deployment scenario.
One-Line Installer for macOS and Linux
Open a terminal and run:
curl -fsSL https://pinchtab.com/install.sh | bashThe script downloads the latest binary, places it in /usr/local/bin, and verifies the installation. ARM64 support is automatic.
Global Installation via npm
For users who prefer npm workflows:
npm install -g pinchtabThis method works on macOS, Linux, and Windows Subsystem for Linux.
Docker Deployment for Production and Self-Hosting
The official container image provides the fastest path to isolated, reproducible environments:
docker run -d \
-p 9867:9867 \
-v ~/.pinchtab:/root/.pinchtab \
--name pinchtab \
pinchtab/pinchtabMount the profile directory to persist sessions across container restarts. Add environment variables for port or authentication token as needed.
Building from Source (Optional)
Clone the repository and run:
git clone https://github.com/pinchtab/pinchtab.git
cd pinchtab
./doctor.sh
go build ./cmd/pinchtabThe resulting binary runs on any platform with a Chrome installation.
After installation, verify with:
pinchtab --versionConfiguring and Launching the Orchestrator
PinchTab reads configuration primarily through environment variables. Start the server with:
pinchtabIt listens on http://localhost:9867 by default. Useful environment variables include:
BRIDGE_PORT— Change the listening port (default 9867)BRIDGE_TOKEN— Require authentication on all API callsCHROME_PATH— Override automatic Chromium detectionHEADLESS— Default launch mode for new instances
Open http://localhost:9867 in a browser to access the real-time dashboard. The interface displays running instances, their status, active tabs, profile disk usage, and quick-action buttons for common operations.
Creating and Managing Browser Instances
The orchestrator treats each Chrome process as an “instance.” Create one with the CLI:
pinchtab instance launchFor a visible window:
pinchtab instance launch --mode headedVia HTTP API:
curl -X POST http://localhost:9867/instances/launch \
-H "Content-Type: application/json" \
-d '{"mode": "headed"}'Each instance receives a unique ID (inst_XXXXXXXX) and an automatically assigned port in the 9868–9968 range. Profiles are assigned or created on demand.
To launch with a named persistent profile:
pinchtab profile create work
pinchtab instance launch --profile workList all instances and their status directly from the dashboard or:
curl http://localhost:9867/instancesInteracting with Tabs and Performing Actions
Every instance can open multiple tabs. Navigate and interact using stable references or URLs.
Basic navigation:
pinchtab nav https://addrom.comCapture the accessibility snapshot (the primary interface for agents):
pinchtab snap -i -cClick an element:
pinchtab click e5Fill a form field:
pinchtab fill e3 "[email protected]"Extract clean text (token-efficient):
pinchtab textThe same operations are available through the HTTP API, returning structured JSON perfect for LLM consumption.
Advanced Multi-Instance Orchestration Workflows
PinchTab shines when scaling to multiple isolated sessions. Launch parallel instances for different accounts or tasks:
for user in alice bob charlie; do
pinchtab instance launch --profile $user --mode headless
doneEach instance operates independently. Agents can send targeted requests to specific instance IDs or ports. The real-time dashboard visualizes resource usage and lets operators pause, resume, or terminate instances instantly.
For agent frameworks supporting the Model Context Protocol, the included SMCP plugin exposes 15 ready-to-use tools (navigate, snapshot, action, etc.) with zero additional dependencies.
Security and Best Practices with Stealth Mode
PinchTab’s stealth injection activates automatically on every new instance. It modifies navigator properties, randomizes fingerprints, and suppresses automation signals before any page loads. Combined with real Chrome execution, this approach bypasses most commercial anti-bot systems.
Security remains the operator’s responsibility. Always set BRIDGE_TOKEN in production. Treat the ~/.pinchtab directory as sensitive data. Start with low-risk accounts and monitor the dashboard for unexpected activity. Because all communication stays on localhost by default, no external data leaves your machine unless you explicitly expose the port.
Real-World Use Cases and Performance Gains
Developers use PinchTab for:
- Autonomous e-commerce agents that maintain login sessions across days
- Parallel lead-generation bots scraping multiple regional sites
- Compliance monitoring agents that export PDFs nightly
- Research agents performing multi-tab analysis on academic portals
Early benchmarks show accessibility-tree snapshots consume 800–1,200 tokens versus 8,000–12,000 for screenshots. Multi-instance orchestration allows a single server to drive dozens of concurrent Chrome processes with minimal overhead.
Getting Started Today
PinchTab transforms browser automation from a fragile, expensive bottleneck into a fast, stealthy, and observable foundation for AI agents. Its tiny footprint, native Docker support, and open MIT license make it ideal for both local experimentation and production self-hosted deployments.
Install the binary now, open the dashboard, and give your agents the real Chrome control they deserve. Full documentation lives at pinchtab.com/docs, and the community repository welcomes contributions for new agent integrations and platform support.
