Tactab

What Tactab MCP server does

The Tactab MCP server gives MCP-compatible clients access to an active Google Chrome tab. It combines visual capture with browser interaction, allowing an agent to inspect page content, retrieve structured information, and perform actions through a single automate_website tool.

Supported operations include taking a high-resolution PNG screenshot, scraping page metadata and visible text, extracting matching elements, reading HTML, clicking controls, filling form fields, scrolling, and navigating to a URL. These capabilities support tasks such as checking a page layout, collecting links, reading pricing content, or entering test data into a form.

The project is intended for the Browser Automation category and runs locally rather than through a hosted automation service. The browser extension can access authenticated tabs and local development sites because the bridge remains on the same machine.

How it works

An MCP client starts the Node.js server through the standard input/output transport. The server translates tool requests into JSON messages sent over a WebSocket connection at ws://127.0.0.1:8765.

The Chrome extension’s Manifest V3 service worker receives those messages and routes them to the active tab. Its content script performs the requested DOM operation or capture, then sends the result back through the extension, WebSocket bridge, and MCP server to the AI client.

This architecture requires both the local server and the unpacked Chrome extension to be running. The extension badge shows green ON when connected and red OFF when disconnected; it retries the connection every five seconds.

Setup and configuration

Install the server dependencies from the repository’s server/ directory with npm install. Node.js 18 or newer is indicated by the project documentation.

Load the extension/ directory into Chrome from chrome://extensions/ after enabling Developer mode and choosing Load unpacked. Then add the server to the MCP configuration for the client you use. The configuration runs node with an absolute path to server/bridge.js.

Claude Desktop and Cursor use the same basic JSON structure. Windows paths in JSON require escaped backslashes. No environment variables are documented as required.

Tools and capabilities

The Tactab MCP server exposes the following actions through automate_website:

• take_screenshot: captures a visual PNG of the active tab.
• scrape_data: returns the page title, URL, meta description, and cleaned inner text; maxLength is optional.
• click_button: targets an element with a CSS selector and can fall back to matching its inner text.
• fill_form: sets an input or textarea value and dispatches input and change events for common frontend frameworks.
• get_elements: returns text, links, classes, and IDs for elements matching a selector, with configurable limits.
• scroll: moves the page up, down, to the top, or to the bottom, with an optional pixel amount.
• navigate: opens a supplied URL in the active tab.
• get_html: returns outer HTML for a selected element or the page body, with an optional length limit.

Limitations and notes

The Tactab MCP server is specific to Google Chrome and relies on its extension being loaded and connected. Actions operate on the active tab and are described for standard webpages. The bridge binds to the local loopback address rather than providing remote browser control.

The repository describes the project as local and private: browser cookies, authenticated tabs, and local development pages are not sent to a cloud automation provider by the bridge. As with any browser automation tool, agents should be given only the tab access and credentials appropriate for the task.