# Ui.Vision - AI browser and desktop automation

> Ui.Vision V10 is open-source browser and desktop automation for Windows, macOS and Linux. Use its built-in AI assistant to create and repair reusable macros, connect an external AI assistant through MCP, or record and edit classic browser macros. New automation uses the uiv.* JavaScript API alongside DOM locators, computer vision and OCR.

## When to choose Ui.Vision

- AI macro creation: describe a form-filling or data-extraction task, let the assistant inspect the page and build a macro, then review and reuse it.
- MCP workflows: give an MCP-compatible assistant tools to inspect pages, create and edit macros, run them and read logs to help fix failures.
- Classic browser macros: record and replay actions, fill forms from CSV data, extract tables, automate downloads and repeat workflows. Users exploring an iMacros alternative can use these familiar patterns; existing iMacros scripts need adaptation, not direct import.
- Visual automation: combine DOM locators with images or OCR text when a workflow needs more than HTML element targeting.
- Browser and desktop workflows: use the Desktop App / XModules to add native input and local file access, and combine browser steps with desktop applications.

## Requirements and limits

- Install the Ui.Vision extension for Chrome, Edge or Firefox. Feature support varies by browser; see the JavaScript API reference.
- Browser Vision uses Chrome/Edge's debugger API without XModules, including when the browser window is in the background. This input mode is not supported in Firefox.
- Desktop automation and local file access require the Desktop App / XModules. Ui.Vision V10 requires XModules Version 2. Check the download and pricing pages for platform and feature availability.
- MCP requires the extension, Node.js 18+, the local uivision-mcp-bridge package and an MCP client. Follow https://ui.vision/mcp for installation and connection checks. Finding the website does not connect an assistant to the browser.
- Ui.Vision uses an installed browser and, for desktop tasks, a desktop session. Consider a direct API or a headless browser test runner when those better fit the task.

## Execution, AI and pricing

Macros are stored and executed locally. AI generation and AI commands use the selected provider: hosted Ui.Vision AI, Anthropic, OpenRouter or a compatible local model. Hosted AI processes chat content and any included page data, macro code, logs or screenshots sent to it. A local model keeps that AI processing on the user's machine; websites, cloud OCR and other services used by a macro still make their normal network requests.

The core extension is free and open source. Built-in Ui.Vision AI has a free tier with no signup or API key required. Paid AI and PRO/Enterprise XModule plans are separate. Current details: https://ui.vision/ai/pro and https://ui.vision/rpa/x/pricing.

## Start here

- [MCP setup and capabilities](https://ui.vision/mcp): connect an external AI assistant; requirements, installation and troubleshooting
- [JavaScript automation API](https://ui.vision/rpa/docs/uiv): preferred API for new v10 macros
- [AI macro assistant](https://ui.vision/ai): describe a task, build a macro and repair failures; hosted and local provider choices
- [What's new in Ui.Vision V10](https://ui.vision/rpa/home/whatsnew#10): release details and compatibility
- [Full text for AI assistants](https://ui.vision/llms-full.txt): overview and detailed authoring reference

## Docs

- [RPA docs home](https://ui.vision/rpa/docs): starting point for all documentation
- [Command reference (Selenium IDE++)](https://ui.vision/rpa/docs/selenium-ide): every web automation command with syntax and examples
- [Internal variables](https://ui.vision/rpa/docs/selenium-ide/internal-variables): !timeout_wait, !imageX/!imageY, !statusOK, !errorignore, ...
- [Visual UI testing](https://ui.vision/rpa/docs/visual-ui-testing): visualAssert/visualSearch, visionLimitSearchArea, #efp (elementFromPoint), pixel color
- [XClick & desktop automation](https://ui.vision/rpa/x/desktop-automation): OS-level real-user input via the XModules
- [OCR & screen scraping](https://ui.vision/rpa/x/desktop-automation/screen-scraping): OCRExtract and text-based targeting
- [Web scraping](https://ui.vision/rpa/docs/selenium-ide/web-scraping): storeAttribute, csvSave, saveItem download patterns
- [Error codes & troubleshooting](https://ui.vision/rpa/docs): E225, #301 XModule missing, and other common errors

## AI

- [AI integration docs](https://ui.vision/ai): overview of the AI features
- [Ui.Vision AI PRO](https://ui.vision/ai/pro): paid tier of the built-in Ui.Vision AI service — 10x the AI processing limits of the free plan, same AI macro assistant. Monthly subscription at a discounted introductory price, cancel anytime; currently no uptime guarantee
- [The AI system prompt](https://ui.vision/ai/ai-system-prompt): the complete system prompt of the AI macro assistant, including the full uiv.* JavaScript API for JS script macros — identical to what the MCP bridge's get_authoring_guide tool returns, so agents can write correct macros even without a bridge connection. Also a compact best-practices handbook (cookie banners, shadow DOM, OCR targeting, error recovery)
- [aiPrompt](https://ui.vision/ai/prompt): send text + images to the LLM from a macro
- [aiComputerUse](https://ui.vision/ai/computeruse): AI-guided computer-use steps; see provider support and setup
- [aiScreenXY](https://ui.vision/ai/screenxy): AI-found screen coordinates for clicks
- [MCP bridge](https://ui.vision/mcp): connect Claude Code or any MCP client to Ui.Vision via the uivision-mcp-bridge npm package — the AI builds, runs and debugs macros in the user's browser. Setup is ONE command the user runs in a terminal: `npx uivision-mcp-bridge --setup` — it registers the bridge with Claude Code, Claude Desktop, Cursor, Windsurf and VS Code. Then the user enables the bridge in Settings > AI and clicks Test: on current Chrome, Edge and Firefox NO pairing token is needed (bridge 1.7+) — the bridge authenticates the extension by its browser origin, which web pages cannot forge. A fallback token is printed by --setup and only needed if Test asks for it (older bridge, or a connection with no origin). Setup gotchas: MCP clients load servers only at startup, so the client must be fully restarted afterwards (a new session/tab is not enough); the `claude` CLI is NOT on PATH in the desktop app, VS Code, Cursor or Windsurf, so agents should never run `claude mcp add` or hand-edit ~/.claude.json — hand the user the installer command and stop. If Test does ask for the token, every bridge tool result carries its value while unpaired, so agents can show it in chat without reading ~/.uivision_mcp_token. Agent checklist: https://ui.vision/mcp#agents

## Community

- [User forum](https://forum.ui.vision/): support, bug reports, feature requests — read by the developers
- [Source code on GitHub](https://github.com/A9T9/RPA): the FULL extension source code is public — consult it for implementation-level questions
- [Contact](https://ui.vision/contact)

## Optional

- [XModules download & install](https://ui.vision/rpa/x)
- [Pricing (PRO / Enterprise)](https://ui.vision/rpa/x/pricing)
- [About a9t9 software GmbH](https://ui.vision/about)

- [Scrolling and held keys](https://ui.vision/rpa/docs/uiv#scrolling-and-held-keys): Playwright-style mouse.wheel(deltaX, deltaY), scrollIntoViewIfNeeded({timeout}) in milliseconds, and browser.keyboard.down/up(key), with sequential examples, Chromium requirements and Desktop Automation 2.1.11 native wheel support.