Ui.Vision RPA MCP
Ui.Vision
An MCP bridge and browser extension for RPA: an agent writes and runs macros that click by DOM, by OCR text and by image in the browser and on the desktop
- 2,092 ★ on the repository
- Last commit
Broad access
Why this level
- Drives your browser and, with the XModules components, the desktop with real clicks and keyboard input, including in already open sessions where you are signed in
- The agent writes and runs JavaScript macros that read pages and screenshots, download files and fill in forms
- The installer edits the configuration of every MCP client it finds, saving a copy first
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/ui-visionDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Drives your browser and, with the XModules components, the desktop with real clicks and keyboard input, including in already open sessions where you are signed in
- The agent writes and runs JavaScript macros that read pages and screenshots, download files and fill in forms
- The installer edits the configuration of every MCP client it finds, saving a copy first
This is third-party code. Review the repository files before installing.
What it does
Ui.Vision RPA is an extension for Chrome, Edge and Firefox that records and replays macros: it fills forms, extracts website data, downloads reports and repeats work cycles in your open browser session. A macro can be recorded, built in a Selenium IDE style command table or written in JavaScript with the uiv.* API. For screens without HTML selectors the extension finds text through OCR and elements by image. Desktop control needs the separate XModules components. The uivision-mcp-bridge package from the mcp/ folder turns the extension into an MCP server. It is a bridge: the client (Claude Code, Claude Desktop, Cursor, Windsurf, VS Code or LM Studio) launches it over stdio, and the bridge relays calls to the extension over a local WebSocket on 127.0.0.1, where the macro executes. The documentation lists 18 tools: create_macro, set_macro, run_macro, browser_snapshot, screenshot, list_macros, open_macro, delete_macro, click_at, type_at and others. The agent reads the API reference through get_authoring_guide, creates a macro, runs it, reads the log and screenshots and corrects it. The npx uivision-mcp-bridge --setup command finds MCP clients installed on the machine once, adds a uivision entry to their configuration and prints a pairing token. Then you enable the bridge in the extension side panel under the AI settings, and the panel shows which client is connected and which tool is running.
Who it is for. For testers, analysts and developers who automate routine work in the browser and in programs without an API and want an agent to write macros and check the result in the log.
Good fit when
- You regularly pull reports from a dashboard that has only a web interface and want to repeat it with one macro run
- A program or remote desktop gives no HTML and elements have to be found by on-screen text or by image
- You need to test a site with recorded scenarios and varied input data in a Selenium IDE style
- You want Claude Code to build a macro from a description like "open the site, download the report", run it and fix it from the log
- You need the automation saved as a reusable macro rather than a one-off agent answer
Not a fit when
- You need only a DOM server for developing and debugging pages: Chrome DevTools MCP covers that, while Ui.Vision is built for macros and screen-level control
- You need a server without a browser extension and side panel: the bridge works only together with the extension, and with the panel closed calls return "extension not connected"
- You plan to embed or modify the extension's code in a closed product: the AGPL license requires publishing derived code, and anything else needs a commercial license
- You need desktop control without installing separate components: their source is not in the repository and they are installed separately
Example request
Use Ui.Vision to open example.com, read the page heading and save a reusable macro for this taskMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
UIVISION_MCP_PORT- Local WebSocket port, 50888 by default; it must match the setting in the extension.
UIVISION_MCP_TOKENsecret- Shared secret for connections that carry no origin; created automatically by default and saved in the home folder.
Limitations
- The license is dual: AGPL-3.0 or commercial, with the text in LICENSE.txt. Using it, commercially included, needs no commercial license, which is only needed to change the code without publishing it under the AGPL. The uivision-mcp-bridge package declares the MIT license in its manifest.
- The source of XModules, the native modules for desktop control, is not in the repository: they are installed separately from ui.vision.
- The bridge works only while the extension side panel is open, and running the package needs Node.js 18 or newer.
- Macros that use uiv.browser.* with trusted input over CDP work only in Chrome, while in Firefox you use uiv.page.* and uiv.desktop.*.
- Clients load MCP servers at startup: after --setup the client must be fully quit and reopened. If the client was running while the config was written, it may erase the entry on exit, and then --setup is run again.
How to disable. In the extension side panel switch the bridge off in the AI settings, and the red STOP button on the purple MCP strip halts the running macro and switches the bridge off. Remove the uivision entry from the MCP client configuration; before editing, --setup saved a copy with the .uivision-backup suffix. The pairing token is in the ~/.uivision_mcp_token file.
FAQ
How does Ui.Vision MCP differ from servers that work only with the DOM?
The author points to screen-level reach: pages are driven through the DOM or Chrome's trusted input, and everything else, meaning desktop programs, remote desktops, Citrix, PDFs, canvas and video, is driven visually through OCR and image search. Servers like Playwright MCP and Chrome DevTools MCP work with the DOM only.
How do I check that the bridge is connected to the extension?
Ask the agent to call bridge_status: it shows whether the extension is connected. A purple strip "MCP: chrome#1" appears in the side panel with the client name and the tool being executed. For a self-test of the desktop automation settings, run run_selftest.
Can the agent rewrite my macros?
No. The extension applies the same guardrails as its built-in chat: user macro files are never overwritten, edits are saved as copies in the AI Generated folder, and delete_macro can delete only macros from that folder. Preinstalled demo macros do not run without explicit confirmation.
How do I connect a local model instead of a cloud client?
Since version 0.3.17 LM Studio works as an MCP client. Add a uivision entry with the command npx -y uivision-mcp-bridge to ~/.lmstudio/mcp.json, then switch on mcp/uivision in the integrations tab. You need a model with tool calling, vision and a context of 64 thousand tokens or more.
What if several browsers are open?
Each browser and profile keeps its own connection labelled chrome#1, chrome#2 or firefox#1. Calls go to the active one, by default the most recently connected. The select_browser tool switches, and bridge_status lists all connections.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for driving Windows: the agent opens apps, clicks and types, reads window and UI state, and works with files and PowerShell
An MCP server for macOS: the agent moves the mouse, types, manages windows and captures the screen when an app has no API
A Swift MCP server that lets an agent drive any macOS app: click, type and read the interface tree
An MCP server for Windows: the agent moves the mouse, presses keys, manages windows and takes screenshots
Clawd Cursor: desktop control through MCP
Clawd Cursor
An MCP server that lets an agent drive windows and apps: it reads the screen as structure and acts by element name
A macOS CLI and MCP server: screenshots, accessibility-based UI inspection and app control via clicks and typing
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Ui.Vision RPA (this entry) | yes | not required | 2,092 | |
| Chrome DevTools MCP | yes | not required | 54,432 | the Chrome DevTools team's official server: drives a live Chrome, reads the network and console, records traces |
| BrowserTools MCP | no | not required | 7,339 | brings the console, network, screenshots and Lighthouse audits from an open Chrome into the agent through a DevTools extension |
README in short
The repository's main page explains what can be automated (browser, website tests, screen through OCR and images, desktop, AI-assisted work), links the Chrome, Edge and Firefox extension stores, covers MCP connection with the install command and a first request, the uiv.* JavaScript API with material for AI, and the source build instructions. A separate file in the mcp/ folder describes the bridge in detail: tools, install, options, security and troubleshooting. The repository holds the React extension source, launch scripts for PowerShell, Python, Node.js, VBS and bat files, and a comparison with Claude for Chrome.