Shell and coding agent on mcp clients
Configuration
View docs{
"mcpServers": {
"rusiaaman-wcgw": {
"command": "uvx",
"args": [
"wcgw@latest"
]
}
}
}You can run the wcgw MCP server on your device to give a fully interactive shell and coding environment to your chat application clients. It enables you and the agent to code, build, and run on your local machine with an integrated editor experience and multiplex terminal support for productive interactions.
To use wcgw with an MCP client, connect to one of the available MCP server configurations and start an interactive session. You can instruct the agent to execute shell commands, read files, edit files, and run your code. The agent can operate in an attached terminal, share the workspace, and preserve context for task checkpoints or knowledge transfer. You can switch between modes to plan, edit, or run code as needed. If you want to explore a project interactively, attach to the working terminal and issue commands through your MCP client, while the agent guides you through changes and verifications.
Prerequisites vary by your environment. The wcgw setup relies on the uv tool for managing the MCP runtime, and you can run the server via different host commands described below.
Install uv using Homebrew, then configure Claude to use the wcgw MCP server.
Install uv (required):
1) Install uv with Homebrew
2) Create or update claude_desktop_config.json with the wcgw MCP server configuration using uvx
3) Restart Claude to apply the new MCP server configuration
Optional: Force a specific shell by adding a --shell setting in the configuration to point to /bin/bash or another shell path, if you prefer a non-default shell.
{
"mcpServers": {
"wcgw": {
"command": "uvx",
"args": ["wcgw@latest"]
}
}
}To force a specific shell, add the --shell flag to the wcgw configuration.
{
"mcpServers": {
"wcgw": {
"command": "uvx",
"args": ["wcgw@latest", "--shell", "/bin/bash"]
}
}
}If you encounter an uv ENOENT error, ensure uv is installed and accessible from your PATH. Verify uv with which uv and adjust the configuration accordingly. If problems persist, run the following test command to verify the runtime:
uv tool run --python 3.12 wcgwOn Windows using WSL, configure Claude to connect through wsl.exe and uvx for the wcgw server.
{
"mcpServers": {
"wcgw": {
"command": "wsl.exe",
"args": ["uvx", "wcgw@latest"]
}
}
}You can run wcgw via Docker. Use the following MCP configuration to start wcgw inside a container. This example binds a host workspace directory into the container and runs wcgw. Adjust paths as needed.
{
"mcpServers": {
"wcgw": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"--mount",
"type=bind,src=/Users/username/Desktop,dst=/workspace/Desktop",
"wcgw"
]
}
}
}If the editor or terminal extension supports multiplexed terminals, you can attach to the terminal that the AI uses for running commands. You can attach from the terminal view or use the screen session if you have a screen-based workflow. This allows you to monitor and interact with long-running tasks without losing context.
Reset the shell and set up the workspace environment for a clean session.
Execute shell commands with timeout control and the ability to send text or keystrokes to the running command.
Read content from one or more files to inform edits or decisions.
Create new files or write to files that are currently empty.
Edit existing files using search/replace blocks with context-aware matching and tolerance for formatting.
Read image files for display or processing within the workflow.
Save project context and files for knowledge transfer or to resume tasks later.