Clawdcursor
Let an AI agent operate real Windows, macOS and Linux apps by element name
- Pricing
- Open source
- Platforms
- Windows, macOS, Linux
- Runs
- Locally
- Version
- 2.1.0
- Listing updated
- today
Similar MCP servers
About Clawdcursor
Clawdcursor is an open-source MCP server for browser automation. It runs locally on your machine and is started by your AI client.
This server has no model and no API key of its own; your AI makes every decision and it carries them out and checks them. It tries the fastest route into each app: the web page itself through its own Chrome or Edge window, then the accessibility tree that screen readers use, then OCR when there is no tree, and screenshots only last.
After each action it reports confirmed, unverified, refused or failed, and an expect option re-reads the screen to flag a deviation. Compact mode shows six tools, while the full surface has 89 smaller ones. A Stop button and a pause card hand sign-ins or 2FA steps to the human, and no tool call runs while the card is up.
The README is frank about limits. It acts with your OS permissions and has no built-in policy engine, and its own benchmark covers one agent on Windows with 15 tasks, of which 14 were completed. Apps that draw their own screen, such as canvases, games and CAD viewports, need OCR or screenshots. It is MIT licensed.
Install
Run in a terminal
claude mcp add clawdcursor -- npx -y clawdcursorAdd to claude_desktop_config.json
{
"mcpServers": {
"clawdcursor": {
"command": "npx",
"args": [
"-y",
"clawdcursor"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"clawdcursor": {
"command": "npx",
"args": [
"-y",
"clawdcursor"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"clawdcursor": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"clawdcursor"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"clawdcursor": {
"command": "npx",
"args": [
"-y",
"clawdcursor"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"clawdcursor": {
"command": "npx",
"args": [
"-y",
"clawdcursor"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Tools this server provides (6)
Taken from the project documentation; the running server may differ.
- accessibility
- Reads a window's parts by name and acts on them.
- system
- System helpers such as copying text, OCR and pausing for the user.
- window
- Window management.
- browser
- Opens and reads web pages through Playwright.
- computer
- Screenshots and screen-position clicks.
- batch
- Runs several actions together.
What it does
- Acts by element name
- Presses Send because it is called Send, not because of its screen position.
- Result verification
- Reports confirmed, unverified, refused or failed after each action.
- Stop and pause controls
- A Stop button halts every process and a pause card hands sign-ins to you.
- Browser control
- Drives Chrome or Edge through Playwright in its own profile.
- OCR fallback
- Reads on-screen text and clicks visible labels when there is no accessibility tree.
Good for
- Fill in a desktop app that has no API
- Automate a GUI-only workflow with an agent
- Operate a web page while a person handles the 2FA step
- Test a desktop app through its accessibility tree
Pricing
Open source
Free, MIT licence
Clawdcursor FAQ
Does it need an API key?
No. It has no model or key of its own.
Which operating systems work?
Windows, macOS and Linux, though only Windows is benchmarked so far.
What can go wrong?
It acts with your permissions and has no built-in policy engine, so you should watch what the agent does.
How many tools does it expose?
Six in compact mode, or 89 smaller tools in full mode.
Is Clawdcursor free?
Yes. Clawdcursor is open source, so you can use and self-host it for free.
How do I add Clawdcursor to Claude Code?
Run this command in your terminal: claude mcp add clawdcursor -- npx -y clawdcursor
What are the best alternatives to Clawdcursor?
The closest alternatives to Clawdcursor are REA, Browser Use, Scrapling MCP Server, Context7 and Chrome DevTools MCP.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.


