Description
Letting AI agents browse usually means stuffing whole HTML pages or screenshots into the model, burning tokens and time. PinchTab is a standalone HTTP server that gives AI agents direct control over Chrome, shipped as a small Go binary of about 12MB.
It turns pages into agent-friendly structured content and clickable element references, so agents navigate, extract and interact with few tokens. Installed as a daemon, it lets agent tools share one browser control plane, with a live dashboard for every instance.
HTTP and MCP:Drive the browser through an HTTP API, MCP or CLI.
Token-efficient:Structured text and element refs instead of raw HTML or images.
Multi-instance:Orchestrate several headless or headed Chrome instances.
Profiles:Named profiles with logins, cookies and extensions that agents pick as needed.
Site audits:Screenshots, console errors, broken assets, accessibility, Core Web Vitals and visual comparison of two versions.
Dashboard:Live view of tabs, memory and instance status.
Local-first security:Loopback-only by default, sensitive endpoints off and browsing limited to local sites until you allow more.
It turns pages into agent-friendly structured content and clickable element references, so agents navigate, extract and interact with few tokens. Installed as a daemon, it lets agent tools share one browser control plane, with a live dashboard for every instance.
Features
HTTP and MCP:Drive the browser through an HTTP API, MCP or CLI.
Token-efficient:Structured text and element refs instead of raw HTML or images.
Multi-instance:Orchestrate several headless or headed Chrome instances.
Profiles:Named profiles with logins, cookies and extensions that agents pick as needed.
Site audits:Screenshots, console errors, broken assets, accessibility, Core Web Vitals and visual comparison of two versions.
Dashboard:Live view of tabs, memory and instance status.
Local-first security:Loopback-only by default, sensitive endpoints off and browsing limited to local sites until you allow more.
