PinchTab

// Overview

PinchTab is a standalone HTTP server that gives AI agents direct control over Chrome. Rather than bundling a browser, it ships as a single self-contained binary of about 30MB that drives a locally installed Chrome or Chromium, and it is designed to be installed as a user-level daemon so several agent tools can reuse one browser control plane running in the background. Agents point at `http://localhost:9867` and get browser tools for navigating pages, extracting structured content and interacting with the DOM.

The project's pitch is token efficiency and stable references. It reports roughly 800 tokens per page using text extraction — which it describes as five to thirteen times cheaper than screenshots — so an agent is not spending its context on raw HTML or images. Its element references are accessibility-first and denote a DOM node rather than a row in a list: the same `e5` survives a change of filter, selector or depth, and expires only when the browser navigates to a new document. That is the difference between an agent that can take three steps on a page and one that loses its grip after the first re-render.

Beyond single-page automation it manages profiles and instances. Headed Chrome profiles can carry authentication, cookies, extensions and a name and description, so a request can select the right profile for a task; multiple Chrome instances can run in parallel with isolated profiles, in containers, or on remote machines. There are also site-level commands: `pinchtab audit` walks a site or its sitemap and produces a report with screenshots, console errors, broken assets, an accessibility score, Core Web Vitals and security findings, while `pinchtab compare` diffs two deployments visually and can fail a CI job on a difference.

PinchTab is explicit that it is a privileged operator tool, and it defaults to a local-first security posture: the server binds to 127.0.0.1, sensitive endpoint families and `attach` are disabled by default, and its IDPI layer restricts browsing to locally hosted sites until you widen the allowlist deliberately. The documentation treats remote, container and distributed deployments as advanced setups you are responsible for securing, and warns that the dashboard, HTTP API and MCP server are not built for untrusted or multi-tenant exposure. It is MIT-licensed, written in Go, and has agent plugins for Grok Build and OpenClaw.

// What it does

• Single self-contained ~30MB Go binary that drives a locally installed Chrome or Chromium
• Control from the CLI, plain curl against the HTTP API, or an MCP server
• Token-efficient extraction — roughly 800 tokens per page, which the project puts at 5–13x cheaper than screenshots
• Accessibility-first element refs that denote a DOM node, so a ref survives filter, selector and depth changes and expires only on navigation
• Headless and headed modes, with named Chrome profiles carrying auth, cookies and extensions
• Multi-instance orchestration across containers or remote machines, plus an advanced attach mode for externally managed Chrome
• Site audits (screenshots, console errors, broken assets, accessibility score, Core Web Vitals, security findings) and visual comparison with a --fail-on-diff CI gate
• Local-first security defaults: loopback bind, sensitive endpoint families off, and browsing restricted to local sites until you allow more
• ARM64-optimized with first-class Raspberry Pi support and automatic Chromium detection

// Getting started

The primary path is to install the binary, install and start the daemon, then point an agent at the local control plane. Running bare `pinchtab` performs first-run security setup and prints server status and next steps.

curl -fsSL https://pinchtab.com/install.sh | bash

pinchtab daemon install

pinchtab server

http://localhost:9867

pinchtab audit https://example.com --output-dir ./audit

pinchtab audit https://example.com/sitemap.xml --sitemap --sample-size 2 --output-dir ./audit

pinchtab compare https://example.com https://staging.example.com --fail-on-diff

grok plugin install pinchtab --trust

openclaw plugins install @pinchtab/pinchtab

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Giving a coding or research agent a persistent local browser instead of spawning a fresh headless session per task
• Logged-in automation via named headed profiles — 'use my work profile and download the weekly report'
• Pre-release site audits and visual regression gates in CI with pinchtab audit and pinchtab compare
• QA and distributed browsing across several parallel Chrome instances with isolated profiles
• Sandboxing browser automation by running the whole control plane inside Docker

// How PinchTab compares

PinchTab alongside other open-source computer & browser use tools AI/TLDR tracks, ranked by GitHub stars.