
Ui.Vision MCP Server: Setup for Claude Code, Cursor and Other Apps
Lets Claude, Cursor and other assistants write, run and fix Ui.Vision browser and desktop automations.
Ui.Vision MCP is made by the Ui.Vision RPA team and links your assistant to the Ui.Vision browser extension. The assistant can look at a page, write a macro, run it, and read the log and screenshots to fix it. It also reaches the desktop through OCR and image search when the desktop add-on is installed.
It runs on your computer. A small helper starts from your assistant and talks to the extension over a local connection on your own machine. You need the Ui.Vision extension installed and the bridge switched on in Settings, then AI, before any tool works. Keep the Ui.Vision side panel open while the assistant works.
The assistant controls a real browser with your logged-in sessions and can click, type and run scripts. There is no read-only mode, so watch the first runs.
Good for
- People who automate web forms, reports and downloads and want an assistant to write the macro
- Testers who replay browser checks and want help fixing broken steps
- Teams that automate desktop or remote-desktop software through screen reading
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install Ui.Vision MCP
Claude Code
- 1
Copy this prompt.
Add the Ui.Vision MCP server to Claude Code. Run this command: claude mcp add ui-vision -- npx -y uivision-mcp-bridge Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Node.js, Ui.Vision browser extension with the MCP bridge enabled (Settings, then AI).
Read-only mode: There is no read-only mode. Use the STOP button in the Ui.Vision panel to halt a run.
Install the Ui.Vision extension from your browser's store first. Then run the setup command from the box once, or use the config above. Open Ui.Vision, go to Settings, then AI, turn on the MCP bridge, keep the port at 50888 and click Test. Quit your assistant completely and reopen it so it loads the server. Keep the side panel open: with it closed, tools answer 'extension not connected'. Desktop automation needs the extra Ui.Vision desktop components.
Built from the package data we hold. Not yet tested by us.
Tools
Checks that the Ui.Vision extension is connected
Tells you whether the Ui.Vision side panel is open and linked to the bridge, and lists each connected browser. Call it first when other tools fail.
Try: “Is Ui.Vision connected to my browser?”
Read ·
bridge_statusShows what the macro runner is doing right now
Reports whether a macro is running, which one is open, and the latest log lines. It answers even while a long run is in progress.
Try: “What is the macro doing right now?”
Read ·
run_statusPicks which connected browser gets the commands
When several browsers are linked, this chooses which one the other tools act on, by label such as chrome#2 or by browser family.
Try: “Switch Ui.Vision to my Firefox browser.”
Write ·
select_browserOpens the Ui.Vision panel in a closed browser
Wakes a browser whose Ui.Vision panel is closed and opens it in a tab, even when the browser has no windows. It is the only way the panel opens without a click from you.
Try: “Open the Ui.Vision panel in Chrome.”
Runs code ·
open_panelLists the saved macros
Returns the name and id of every macro stored in Ui.Vision.
Try: “List my Ui.Vision macros.”
Read ·
list_macrosOpens a saved macro in the editor
Loads the named macro into the editor and returns its content. Unsaved changes in the editor can be lost.
Try: “Open the macro "invoice download".”
Read ·
open_macroReturns the guide for writing macros
Returns the reference for the uiv.* JavaScript commands, the classic command table and locator rules. The assistant reads it once before writing its first macro.
Try: “Read the Ui.Vision macro guide before you start.”
Read ·
get_authoring_guideShows the macro open in the editor
Returns the macro that is currently open, as Ui.Vision JSON.
Try: “Show me the macro that is open now.”
Read ·
get_macroSends a message to the built-in Ui.Vision AI chat
Types a message into Ui.Vision's own AI chat, which then runs it with its own tools. It returns at once and the chat can take minutes.
Try: “Ask the Ui.Vision chat to open example.com and read the heading.”
Runs code ·
send_chatReads the built-in AI chat
Returns the text of the Ui.Vision chat and whether a run is still going. Use it after send_chat.
Try: “Is the Ui.Vision chat done yet?”
Read ·
get_chatReloads the Ui.Vision extension
Reloads the extension from disk, like the reload button on the browser extensions page. The panel reconnects by itself.
Try: “Reload the Ui.Vision extension.”
Runs code ·
reload_extensionRuns the desktop automation self-tests
Runs all the desktop automation checks in Settings: native host, image search, screenshots, mouse input and file access. It moves the real mouse while it runs and returns one report.
Try: “Run the Ui.Vision self-test.”
Runs code ·
run_selftestShows the AI provider settings
Reports the active AI provider and model, which providers have keys stored (yes or no only), and the OCR engine. Keys are never shown.
Try: “Which AI provider is Ui.Vision using?”
Read ·
get_ai_settingsSwitches the AI provider settings
Changes the AI provider, model, vision scope or OCR engine among those you already set up. It cannot set or read API keys.
Try: “Switch Ui.Vision to the anthropic provider.”
Write ·
set_ai_settingsEdits the macro in the editor
Replaces the open macro with the JSON you give it. A macro of yours is never overwritten: the first change is saved as a copy in the "AI Generated" folder.
Try: “Fix the login step in this macro.”
Write ·
set_macroCreates a new macro
Saves a new macro under a unique name in the "AI Generated" folder and opens it in the editor.
Try: “Create a macro that logs into example.com and downloads the report.”
Write ·
create_macroDeletes a macro from the "AI Generated" folder
Deletes one macro, and only from the "AI Generated" folder. Anything outside that folder is refused.
Try: “Delete the scratch macros you made.”
Delete ·
delete_macroRuns a macro or a script in your browser
Runs a saved macro, or JavaScript you give it, in the browser tab and waits until it ends. It returns the log and a marked picture of the page. It clicks and types for real, so what it does can have effects.
Try: “Run the "invoice download" macro.”
Runs code ·
run_macroReads the structure of the current page
Returns the page as a list of buttons, fields and links, each with a reference the macro can click or fill. Can open a URL first.
Try: “Show me the form fields on this page.”
Network ·
browser_snapshotOlder name for browser_snapshot
An alias of browser_snapshot kept for older prompts. It behaves the same.
Try: “Get the page structure.”
Network ·
get_pageClicks a point on the last screenshot
Clicks the point you name on the most recent screenshot, in the browser tab or on the desktop. It is meant for exploring, not for saved macros.
Try: “Click the Save button in the screenshot.”
Runs code ·
click_atClicks a point and types text
Clicks a point on the latest screenshot and types text there. Key names such as Enter and Tab are honoured.
Try: “Click the search box and type "invoices".”
Runs code ·
type_atTakes a screenshot of the browser or the screen
Captures the visible tab, or the whole screen when you ask for the desktop. It can zoom into a region or number the clickable elements.
Try: “Take a screenshot of the page.”
Read ·
screenshotSaves a crop as a reference image
Crops a box from the latest screenshot and stores it as an image the macro can find on screen later. Saving the same name overwrites the old image.
Try: “Save the logo as an image called logo.”
Write ·
save_element_imageSaves an anchor and click target as an image
Saves an anchor box plus a click offset from the latest screenshot, so a macro can click next to something it recognises.
Try: “Save the label as an anchor and click the field beside it.”
Write ·
save_relative_image
Read from the source code
Permissions
high risk · 4 things to watch
High risk
The assistant can drive your real browser, and the desktop if you add the desktop component, so it can click, type and run scripts as you. Use a separate browser profile without sensitive logins, and watch the first runs.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It runs scripts in your logged-in browser
run_macro runs macros and inline JavaScript with real clicks and typing in your browser, where you may be signed in to email or banking. Use a separate browser profile and press STOP if it goes wrong. github.com
It can control your whole screen
With the desktop add-on, click_at, type_at and run_selftest move the real mouse and type on your desktop. Do not leave other apps open with private data while it runs. github.com
It deletes macros in one folder
delete_macro removes only macros inside "AI Generated". Your own macros are refused, and set_macro saves a copy instead of overwriting them. github.com
It can open your browser
open_panel wakes a closed browser, and the README says the assistant may start the browser from its shell. This only happens when the assistant asks for it. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only mode is documented. Ui.Vision shows a banner while the assistant is in control, and a STOP button in the side panel halts the run and switches the bridge off. |
|---|---|
| Talks to | Any site the macro opens in your browser. Local connection to the Ui.Vision extension (127.0.0.1) |
| Sign-in and keys | No account or API key. The helper listens only on 127.0.0.1, and the extension proves it is the real one through its browser origin. |
Questions
Is the Ui.Vision MCP server official?
Yes. The Ui.Vision RPA team ships it in the same repository as the Ui.Vision extension it controls.
Do I need the Ui.Vision extension?
Yes. The server only relays commands to the Ui.Vision extension, which runs them. Install the extension from your browser's store, turn on the MCP bridge in Settings, then AI, and keep its side panel open.
Which AI apps does it work with?
The README names Claude Code, Claude Desktop, Cursor, Windsurf, VS Code and LM Studio. Run the setup command in the install box once and it registers the server with the apps it finds. For others, add the same npx entry by hand.
Does it need an API key or account?
No. The helper and the extension talk over a local connection on port 50888 by default. A pairing token is only a fallback for older versions.
Can it run in read-only mode?
No. Its main tool runs macros and scripts that click and type for real. The STOP button in the Ui.Vision panel halts a run and switches the bridge off.
Does it send my pages to the cloud?
The helper itself listens on 127.0.0.1 only. Pages are read by your assistant, so their content goes wherever that assistant sends it. Ui.Vision's own chat can use its AI provider if you pick one.
How is it different from Playwright MCP?
The README says Playwright MCP and Chrome DevTools MCP drive the page structure only. This one also reads the screen with OCR and image search, so it can automate desktop apps, remote desktops and PDFs.
Information
updated Nov 2, 2026
- Publisher
- Official
- Package
uivision-mcp-bridge- Registry
- npm
- Transport
- stdio
- Version
- 1.7.18
- License
- MIT
- Last commit
- Oct 28, 2026
- Open issues
- 0
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 3Hide
- github.com · github.com
- github.com · github.com
- github.com · github.com