PowerShell.MCP
MCP server giving an agent a PowerShell 7 console shared with the user: commands appear live, passwords and MFA codes stay manual.
- 110 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent gets full system access as the user, and the author warns of possible severe damage.
- Remote mode opens a TCP port, and agent commands can change cloud resources in a signed-in session.
Install
- Works in
Manual install
Install-PSResource PowerShell.MCPThe Windows Setup command from the README, run in PowerShell 7.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent gets full system access as the user, and the author warns of possible severe damage.
- Remote mode opens a TCP port, and agent commands can change cloud resources in a signed-in session.
This is third-party code. Review the repository files before installing.
What it does
PowerShell.MCP connects an agent to a real PowerShell 7 console on Windows, Linux or macOS. Every command runs in a window where the user sees input and output at once, and the session keeps modules, variables and cloud sign-ins between calls. Through that console the agent drives any command-line tool (git, docker, kubectl, terraform, az, aws) and PowerShell Gallery modules. Password prompts, `Get-Credential` and MFA codes are typed by the person in the console and never reach the agent. There are eleven tools in total, among them `start_console`, `execute_command`, `wait_for_completion` and `cancel`, plus a mode for a console on a remote machine.
Who it is for. Administrators and developers who hand system tasks to an agent and want to see every command it runs.
Good fit when
- An agent builds a report from Azure, AWS or Microsoft 365 inside an already signed-in session.
- Several Claude Code sessions run in parallel, each with its own console.
- Building, signing and pushing a tag needs a passphrase typed by hand.
- The agent commands must be recorded through `Start-Transcript`.
Not a fit when
- Full agent access to the system is unacceptable on a production or shared server.
- PowerShell 7.4 cannot be installed.
- The built-in client tools are enough and no extra console is needed.
Example request
Show the five largest log files in the current folder and say which are safe to delete.MCP
- Transport
- stdio
- Authentication
- not required
Environment variables
POWERSHELL_MCP_TIMEOUT_CEILING- Upper limit for command wait time in seconds.
Limitations
PowerShell 7.4 or newer is required. The module gives the agent full system access as the current user, and the author warns of possible severe damage, so use a trusted environment. On Linux a graphical desktop is needed to see the console. Remote mode opens a TCP port while active, so enable it deliberately. Scripts created by the agent stay on disk and run as you, so review them before running them again.
How to disable. Run `claude mcp remove pwsh -s user` and remove the module with `Uninstall-PSResource PowerShell.MCP`. Clean any Transcript logs separately.
FAQ
Does the agent see typed passwords?
No. Input through `Read-Host -AsSecureString`, `Get-Credential` and MFA prompts goes straight to the program and never appears in the agent output.
Can the built-in client shells be turned off?
Yes. `Register-PwshToClaudeCode` asks during registration, and `-DisableBuiltInShellTools` or `-KeepBuiltInShellTools` answer in advance.
How do I run several agents at once?
Each client instance gets its own console with a unique window title. The `Get-MCPOwner` cmdlet shows who owns the current console.
Can another machine be controlled?
Yes. Run `Enable-MCPRemoteControl` on the remote side and connect with `connect_remote_console` using the pairing code.
Which platforms are supported?
Windows 10 and 11, Windows Server 2016 or newer, Ubuntu 22.04 or newer, Debian 11 or newer, RHEL 8 or newer and macOS 12 or newer. PowerShell 7.4 or newer is required everywhere.
Discussion
No questions or reviews yet.
Related
MCP serversConsole Automation MCP gives an agent local shells and SSH sessions: input, output, error detection, long processes
MCP SSH Agent is an MCP server that reads hosts from ~/.ssh/config, runs commands and copies files through system ssh and scp.
A WebSSH console for VPS with an MCP server: the agent works across several servers while ACL tokens limit tools and hosts.
An MCP server giving ChatGPT and agents a terminal, files, Playwright and remote machines inside an isolated workspace.
Muxin MCP connects an agent to your computers, with remote sessions that keep running after the application window closes.
A secure MCP runtime through which an agent operates local machines, servers and containers
README in short
The agent and the person share one console: commands are visible at once, the session persists and secrets stay with the human.