ssh-winrm-mcp
MCP server for administering Linux/Unix hosts via SSH and Windows hosts via WinRM/PowerShell Remoting, supporting persistent inventory, sessions, jobs, and command groups.
README
ssh-winrm-mcp
A small, pure-Python MCP server for administering Linux/Unix hosts over SSH and Windows hosts over WinRM/PowerShell Remoting.
It is designed to run directly from GitHub with uvx. No repository checkout and no config path are required. The server starts with an empty inventory and can create and maintain its own persistent local inventory through MCP tools.
Features
- Stored connection profiles or completely ad-hoc connection settings.
- SSH password, private key, SSH agent, host-key verification, commands, programs, scripts, SFTP files, and recursive directories.
- WinRM/PSRP with NTLM, Basic, Negotiate, Kerberos, CredSSP, or certificate settings supported by
pypsrp. - PowerShell and CMD execution, script/program execution, and file or directory transfer.
- Named persistent SSH shells and WinRM PowerShell runspaces.
- Long-running asynchronous jobs inside named sessions.
- Persistent local inventory with initialization, validation, create, read, update, rename, delete, tags, and atomic writes.
- Built-in command groups for system information, services, restart/shutdown, and local user management.
- Persistent external command-group folder for specialized modules.
stdio, Streamable HTTP, and SSE transports.
Run directly from GitHub
After publishing the repository as YOUR_GITHUB_USER/ssh-winrm-mcp:
uvx --from git+https://github.com/YOUR_GITHUB_USER/ssh-winrm-mcp.git ssh-winrm-mcp
For a fixed release or commit:
uvx --from git+https://github.com/YOUR_GITHUB_USER/ssh-winrm-mcp.git@v0.2.0 ssh-winrm-mcp
The default transport is stdio, suitable for a local MCP client.
VS Code MCP configuration
{
"servers": {
"ssh-winrm": {
"type": "stdio",
"command": "uvx",
"args": [
"--from",
"git+https://github.com/YOUR_GITHUB_USER/ssh-winrm-mcp.git",
"ssh-winrm-mcp"
]
}
}
}
No inventory environment variable is required.
Local development
git clone https://github.com/YOUR_GITHUB_USER/ssh-winrm-mcp.git
cd ssh-winrm-mcp
uv sync --extra dev
uv run pytest
uv run ssh-winrm-mcp
Automatic local storage
When SSH_WINRM_MCP_CONFIG is not set, the inventory is stored here:
- Linux/macOS:
${XDG_CONFIG_HOME:-~/.config}/ssh-winrm-mcp/inventory.json - Windows:
%APPDATA%\ssh-winrm-mcp\inventory.json
The file does not need to exist before startup. The first profile create operation also creates it automatically. Writes are atomic and the file is restricted to the current user where the operating system supports Unix permissions.
A custom path remains possible:
SSH_WINRM_MCP_CONFIG=/srv/private/remote-inventory.json \
uvx --from git+https://github.com/YOUR_GITHUB_USER/ssh-winrm-mcp.git ssh-winrm-mcp
Inventory tools
remote_inventory_statusremote_inventory_initremote_inventory_validateremote_profile_listremote_profile_getremote_profile_createremote_profile_updateremote_profile_renameremote_profile_delete
An agent can start with no file and create a profile itself:
remote_inventory_init()
remote_profile_create(
name="linux01",
settings={
"protocol": "ssh",
"host": "linux01.home",
"username": "admin",
"private_key_path": "~/.ssh/id_ed25519",
"verify_host_key": true,
"tags": ["linux", "lab"]
}
)
Update only selected fields:
remote_profile_update(
name="linux01",
changes={"host": "192.168.1.20", "tags": ["linux", "production"]}
)
Remove optional fields:
remote_profile_update(
name="linux01",
changes={},
remove_fields=["password", "password_env"]
)
Profiles may store inline credentials, but that means plaintext secrets are present in the protected JSON file. Prefer password_env, private_key_passphrase_env, and certificate_key_password_env where practical.
See inventory.example.json for SSH and WinRM examples.
Ad-hoc use
Every one-shot operation and remote_session_open accepts a full connection object without saving it:
{
"protocol": "ssh",
"host": "192.168.1.50",
"username": "root",
"password": "temporary-password",
"verify_host_key": false
}
A connection object can also override selected fields from a saved profile.
Main remote tools
One-shot operations
remote_testremote_executeremote_run_programremote_run_scriptremote_uploadremote_download
Example:
remote_execute(profile="linux01", command="uname -a")
remote_run_script(
profile="win01",
language="powershell",
script="Get-Service | Select-Object -First 10"
)
Persistent sessions
remote_session_openremote_session_listremote_session_executeremote_session_closeremote_session_close_allremote_session_cleanup_idle
SSH uses a real reusable remote shell, so cd, shell variables, and exported environment state persist:
remote_session_open(name="linux-maint", profile="linux01", shell="bash")
remote_session_execute(name="linux-maint", command="cd /opt/myapp")
remote_session_execute(name="linux-maint", command="export RELEASE=2026.07")
remote_session_execute(name="linux-maint", command="pwd; echo $RELEASE")
WinRM uses one reusable PSRP PowerShell runspace. Use global variables when state must survive separate PowerShell pipelines:
remote_session_open(name="win-maint", profile="win01", shell="powershell")
remote_session_execute(name="win-maint", command="$global:Target='C:\\Apps'")
remote_session_execute(name="win-maint", command="Get-ChildItem $global:Target")
Sessions are memory-only. Restarting the MCP process closes them. Commands in one session are serialized; create several named sessions for parallel work.
Long-running jobs
remote_job_startremote_job_listremote_job_statusremote_job_resultremote_job_cancel
remote_job_start(
session_name="linux-maint",
command="./upgrade.sh",
timeout=86400
)
The call returns a job ID immediately. Poll with remote_job_status, then retrieve the result with remote_job_result. Current job output is returned when the command finishes; it is not incrementally streamed while running.
Command groups
Built-in groups currently include:
system: information, disk usage, processes, services, reboot, shutdown.users: list, create, delete, enable, disable, and group membership.
Use:
remote_command_groupsremote_command_group_describeremote_command_group_run
The persistent plug-in directory defaults beside the inventory:
~/.config/ssh-winrm-mcp/commands/
On Windows it is under %APPDATA%\ssh-winrm-mcp\commands\.
Create it through MCP:
remote_command_groups(create_directory=true)
Place Python modules there and call remote_command_groups(reload=true). The examples/custom_maintenance.py file is a starting point. A module exports GROUP_NAME and an ACTIONS dictionary containing CommandAction objects.
A custom location may be selected with SSH_WINRM_MCP_COMMANDS_DIR.
HTTP mode
SSH_WINRM_MCP_TRANSPORT=streamable-http \
SSH_WINRM_MCP_HOST=127.0.0.1 \
SSH_WINRM_MCP_PORT=8765 \
uvx --from git+https://github.com/YOUR_GITHUB_USER/ssh-winrm-mcp.git ssh-winrm-mcp
Endpoint:
http://127.0.0.1:8765/mcp
Do not expose this server to an untrusted network. It intentionally provides remote command execution and credential-handling capabilities.
Optional environment variables
SSH_WINRM_MCP_CONFIGSSH_WINRM_MCP_TRANSPORTSSH_WINRM_MCP_HOSTSSH_WINRM_MCP_PORTSSH_WINRM_MCP_LOCAL_ROOTSSH_WINRM_MCP_MAX_OUTPUT_CHARSSSH_WINRM_MCP_COMMANDS_DIR
Project layout
ssh-winrm-mcp/
├── pyproject.toml
├── inventory.example.json
├── src/ssh_winrm_mcp/
│ ├── server.py
│ ├── operations.py
│ ├── sessions.py
│ ├── profiles.py
│ ├── backends/
│ └── commands/
├── examples/
└── tests/
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.