MCP server for PinchTab — control Chrome (or any Chromium browser) via accessibility tree snapshots through the Model Context Protocol.
PinchTab is included as a dependency and starts automatically — no manual setup required.
npx pinchtab-mcpOr install globally:
npm install -g pinchtab-mcp
pinchtab-mcpRelated MCP server: cdpilot
Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"pinchtab": {
"command": "npx",
"args": ["-y", "pinchtab-mcp"]
}
}
}With a custom browser (see Configuration):
{
"mcpServers": {
"pinchtab": {
"command": "npx",
"args": ["-y", "pinchtab-mcp"],
"env": {
"CHROME_BINARY": "/Applications/Yandex.app/Contents/MacOS/Yandex",
"BRIDGE_HEADLESS": "false"
}
}
}
}Cursor / VS Code — add to .cursor/mcp.json or .vscode/mcp.json:
{
"servers": {
"pinchtab": {
"command": "npx",
"args": ["-y", "pinchtab-mcp"]
}
}
}Claude Code:
claude mcp add pinchtab -- npx -y pinchtab-mcpVariable | Default | Description |
|
| PinchTab API endpoint |
| (empty) | Auth token (must match PinchTab's |
| (auto-detect) | Explicit path to PinchTab binary |
Binary lookup order: PINCHTAB_BIN → node_modules/.bin/pinchtab → system PATH.
PinchTab itself is configured via env variables or a config file (~/.config/pinchtab/config.json):
Variable | Default | Description |
| (bundled Chromium) | Path to browser binary — use this for Yandex Browser, Brave, Edge, etc. |
| (none) | Connect to an already-running browser's DevTools (e.g. |
|
| Set to |
|
| PinchTab HTTP API port |
| (none) | Protect the API with an auth token |
|
| Browser profile directory (cookies, sessions) |
|
| Max open tabs |
|
| Block ads |
|
| Block image loading (faster scraping) |
|
| Block video/audio |
|
| Disable CSS animations |
|
| Anti-detection level |
| (system) | Override timezone |
| (auto) | Override User-Agent |
| (none) | Extra flags passed to the browser |
# Yandex Browser
CHROME_BINARY="/Applications/Yandex.app/Contents/MacOS/Yandex" npx pinchtab-mcp
# Brave
CHROME_BINARY="/Applications/Brave Browser.app/Contents/MacOS/Brave Browser" npx pinchtab-mcp
# Microsoft Edge
CHROME_BINARY="/Applications/Microsoft Edge.app/Contents/MacOS/Microsoft Edge" npx pinchtab-mcp
# Visible browser (non-headless) for debugging
BRIDGE_HEADLESS=false npx pinchtab-mcp
# Custom window size (default: random common resolution for stealth)
CHROME_FLAGS="--window-size=1920,1080" npx pinchtab-mcp
# Connect to an existing browser session (keeps your tabs, cookies, logins)
CDP_URL="http://localhost:9222" npx pinchtab-mcpFor tools reference, project structure, agent workflow, and code style see PROJECT_INDEX.md.
See CONTRIBUTING.md.
MIT
Click an element by its ref ID (e.g. 'e5'). Uses human-like click by default. Set waitMs to get a snapshot after clicking (saves a round-trip). For SPAs, clicking may not cause full navigation.
Inputs: ref (required), humanClick, waitMs, waitNav.
Close the current tab or a specific tab by its ID.
Inputs: tabId.
Get all cookies for the current page.
Inputs: no named parameters.
Execute JavaScript code in the page context. Returns the result of the expression.
Inputs: code (required).
Focus an element by its ref ID. Useful for triggering focus-dependent UI (e.g. autocomplete dropdowns) without clicking.
Inputs: ref (required).
Get readable text content from the page (~800 tokens). Best for extracting article content or page information.
Inputs: no named parameters.
Check if PinchTab server is running and responsive.
Inputs: no named parameters.
Hover over an element by its ref ID. Triggers mouseover/mouseenter events — useful for revealing tooltips, dropdown menus, or hover states.
Inputs: ref (required).
List all open browser tabs. Returns tab IDs, URLs, and titles.
Inputs: no named parameters.
Export the current page as a PDF. Returns the PDF as base64.
Inputs: no named parameters.
Press a keyboard key (e.g. 'Enter', 'Tab', 'Escape', 'ArrowDown'). Optionally target a specific element.
Inputs: key (required), ref.
Take a screenshot of the page. Returns a base64-encoded image.
Inputs: no named parameters.
Scroll the page or a specific element. Supports all four directions.
Inputs: direction (required), amount, ref.
Select an option in a dropdown by its ref ID. Pass the option value or visible text.
Inputs: ref (required), value (required).
Get an accessibility tree snapshot of the page. Returns element refs (e0, e1, …) that can be used with click/type/press. Use filter='interactive' for fewer tokens. Use format='compact' for minimal output.
Inputs: diff, filter, format.
Type text into an input field by its ref ID. Uses human-like typing by default. Set clearFirst=true to click, select all, then type — required for React/Vue/Angular inputs where direct fill doesn't trigger state updates.
Inputs: ref (required), text (required), clearFirst, fast, +1 more.
Wait for a specified number of seconds. Useful after navigation or dynamic content loading.
Inputs: seconds.
Wait for a CSS selector to appear on the page. Polls every 500ms up to the timeout. Useful for waiting on dynamic content, modals, or lazy-loaded elements.
Inputs: selector (required), timeoutMs.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 19 tools
Each tool targets a distinct browser automation action (click, type, scroll, navigate, etc.) with no functional overlap. Even similar actions like click vs focus vs hover are clearly differentiated by their descriptions and use cases.
All tools use the pinchtab_ prefix and snake_case. However, there is a minor inconsistency: some tools use verb_noun (e.g., pinchtab_list_instances, pinchtab_get_text) while others are just a noun (e.g., pinchtab_cookies, pinchtab_screenshot) instead of pinchtab_get_cookies or pinchtab_take_screenshot.
19 tools cover the essential browser automation operations without being excessive. Each tool serves a clear purpose, and the number feels well-scoped for a comprehensive automation server.
The tool set covers most common browser interactions: navigation, input, events, screenshots, PDF, cookies, evaluation, and tab management. Missing features like file upload or drag-and-drop are minor gaps that can be worked around with eval.
Automate cloud Chrome—navigate, click, type, screenshot, run code, record screen video
Hosted browser for AI agents: screenshots, post-JS DOM, console, WCAG. No install, no API key.
Cloud browser, web extraction, screenshots, and remote CDP tools for AI agents.
Give AI agents a real cloud Chrome: open sessions, navigate, read pages, click, type and screenshot.