Shows my svg
[![Python](https://img.shields.io/badge/Python-333333?logo=python&logoColor=white&labelColor=333333)](#) [![macOS](https://img.shields.io/badge/macOS-000000?logo=apple&logoColor=F0F0F0)](#) [![Discord](https://img.shields.io/badge/Discord-%235865F2.svg?&logo=discord&logoColor=white)](https://discord.com/invite/mVnXXpdE85) [![PyPI](https://img.shields.io/pypi/v/cua-computer?color=333333)](https://pypi.org/project/cua-computer/)

# cua-mcp-server cua-mcp-server is an MCP server for the Computer-Use Agent (CUA). It enables CUA to run through MCP clients such as Claude Desktop and Cursor. ## Prerequisites - Install lume: https://github.com/trycua/cua/blob/main/libs/lume/README.md#install - Python 3.10+ - pip, venv, setuptools ## Install Download and run the installation script: ```bash curl -fsSL https://raw.githubusercontent.com/trycua/cua/main/libs/python/mcp-server/scripts/install_mcp_server.sh | bash ``` Add this to your MCP client configuration: ```json { "mcpServers": { "cua-agent": { "command": "/usr/bin/env", "args": [ "bash", "-lc", "export CUA_MODEL_NAME='anthropic/claude-3-5-sonnet-20241022'; ~/.cua/start_mcp_server.sh" ] } } } ``` ## Development (run from a local checkout) Use an absolute path to the repository root in the arguments below. ```json { "mcpServers": { "cua-agent": { "command": "/usr/bin/env", "args": [ "bash", "-lc", "export CUA_MODEL_NAME='huggingface-local/ByteDance-Seed/UI-TARS-1.5-7B'; /Users/your-username/Documents/GitHub/cua/libs/python/mcp-server/scripts/start_mcp_server.sh" ] } } } ``` Notes: - Replace `/Users/your-username/Documents/GitHub/cua` with the absolute path to your clone. - The script sets `PYTHONPATH` for local libs and runs the server module. ## Quick Start After configuring your MCP client, restart it and invoke one of these tools: - Take a screenshot ```json { "tool": "screenshot_cua", "args": {} } ``` - Run a task ```json { "tool": "run_cua_task", "args": { "task": "Open Safari and search for University of Toronto" } } ``` Expected results: - Assistant messages streamed during execution - A final screenshot image ## Desktop Extension CUA also provides a Claude Desktop Extension (`.mcpb` file) for easy one-click installation. ### Building the Desktop Extension To build the desktop extension from source: ```bash cd libs/python/mcp-server python3 build-extension.py ``` This creates `desktop-extension/cua-extension.mcpb` which can be installed directly in Claude Desktop. ### Installing the Desktop Extension 1. Build the extension (see above) 2. Open Claude Desktop Settings 3. Go to Extensions 4. Drop the `cua-extension.mcpb` file into the window 5. Follow the installation prompts See [desktop-extension/README.md](desktop-extension/README.md) for more details. ## Documentation - Installation: https://cua.ai/docs/libraries/mcp-server/installation - Configuration: https://cua.ai/docs/libraries/mcp-server/configuration - Usage: https://cua.ai/docs/libraries/mcp-server/usage - Tools: https://cua.ai/docs/libraries/mcp-server/tools - Client Integrations: https://cua.ai/docs/libraries/mcp-server/client-integrations - LLM Integrations: https://cua.ai/docs/libraries/mcp-server/llm-integrations ## Troubleshooting Server reports disconnected in MCP client: - Use an absolute path in the `args` command. - Launch via `/usr/bin/env bash -lc` so the shell initializes and expands paths. - Run the script manually to verify: ```bash /usr/bin/env bash -lc '/Users/your-username/Documents/GitHub/cua/libs/python/mcp-server/scripts/start_mcp_server.sh' ``` pip not found in venv: ```bash python3 -m ensurepip --upgrade python3 -m pip install -U pip setuptools wheel ``` Pydantic schema error related to Image: ```bash python3 -m pip install -U "mcp>=1.2.0" "fastmcp>=0.4.7" "pydantic>=2.7,<2.12" ``` If issues persist, capture logs from your MCP client and the server startup script for diagnosis.