Agent Infra Commands MCP Server: Setup for Claude Code, Cursor and Other Apps
@agent-infra/mcp-server-commandsbytedance/ui-tars-desktop/tree/HEAD/packages/agent-infra/mcp-servers/commandsLets Claude, Cursor and other assistants run shell commands and scripts on your computer.
This server lets Claude, Cursor and other assistants run shell commands and scripts on your computer and read the output. It runs locally with npx and needs no account. It is powerful and has no read-only mode, so it can change or delete anything your user account can. Approve every call.
Good for
- Developers who want the assistant to run builds, tests and git commands
- People setting up a throwaway machine or container for agent experiments
- Users of an assistant that has no built-in shell tool
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 Agent Infra Commands MCP
Claude Code
- 1
Copy this prompt.
Add the Agent Infra Commands MCP server to Claude Code. Run this command: claude mcp add agent-infra-commands -- npx -y @agent-infra/mcp-server-commands 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.
Built from the package data we hold. Not yet tested by us.
Tools
Runs a shell command on your computer
Runs one command with its arguments and returns the output and any errors. You can set the folder it runs in.
Try: “Run `git status` in my project folder.”
Runs code ·
run_commandRuns a script in the interpreter you choose
Pipes a script into an interpreter such as bash, zsh or python and returns the output. You can set the folder it runs in.
Try: “Run a python script that counts the lines in every .md file here.”
Runs code ·
run_script
Read from the source code
Permissions
high risk · 3 things to watch
High risk
The assistant can run any command your user account can run, including commands that delete files or send data over the network, and there is no read-only mode. Approve each call yourself and do not use it with accounts or machines you cannot afford to break.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It runs any command with your permissions
Commands and scripts run through your shell, so they can read, change and delete anything you can. Keep approval prompts on and review each command. github.com
Network mode has no sign-in
With --port the server listens on 127.0.0.1 by default. The source lets you bind it to every interface with --host 0.0.0.0, which would let others on the network run commands. Do not do that. github.com
Output goes straight to the assistant
Whatever a command prints, including secrets from environment files, is returned to the chat. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only mode. Both tools run what they are given with your permissions. |
|---|---|
| Talks to | Whatever the commands you run contact |
| Sign-in and keys | No account or key. It runs with the permissions of the user who starts it. |
Questions
Is it safe to use?
Only with care. It can run any command as you, so a wrong or tricked request can delete files or leak data. Keep approval prompts on and use it in a throwaway folder or machine when you can.
Can I limit where it runs?
You can set a starting folder with the --cwd option, but that is only a default. Commands can still go anywhere your user can.
Does it need an API key?
No. It runs on your computer with no account.
Does it work with Claude Desktop, Cursor and VS Code?
Yes. The README gives setup for VS Code, Cursor, Windsurf and Claude Desktop. It also has a network mode with a local web address.
Where does this code come from?
The source says it is modified from the open-source mcp-server-commands project by g0t4 (MIT) and published by ByteDance under the Agent Infra name.
Information
updated Nov 2, 2026
- Publisher
- Community
- Package
@agent-infra/mcp-server-commands- Registry
- npm
- Transport
- stdio, SSE, Streamable HTTP
- Auth
- None
- Version
- 1.0.0
- License
- Apache-2.0
- Last commit
- Oct 9, 2026
- Open issues
- 358
- 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
- Commands server README · github.com
- Tool registrations · github.com
- Command-line options · github.com