Windows MCP

Community

Windows-MCP enables seamless integration between AI agents and the Windows operating system.

How to install ↓ Compare GitHub source repository ↗

Category: Other #168 of 230
Stars: ★ 8.7k Very popular
Transport: stdio (local process) · SSE · Streamable HTTP
Runtime: Python 3.13+ · Prebuilt binary
Credentials: No credential needed
License: MIT
Last commit: 2 days ago
Tools: 20

47FMRS · D

Windows-MCP is a powerful tool to connect AI agents with Windows, offering a comprehensive set of UI automation tools. Its lightweight design and broad compatibility make it easy to integrate. However, due to system-level privileges, careful security configuration is essential, especially for remote access. Overall, it's a reliable choice for users needing Windows automation.

Strongest · Documentation 14/20 Weakest · Security and permissions 3/20

Reliability

7/20

Security and permissions

3/20

Maintenance

10/20

Documentation

14/20

Setup experience

13/20

Why each score

Reliability 7/20: The README and manifest describe a plausible uvx/stdio MCP server with a rich tool set, so the happy path seems reasonable; however, no source code, CI, or tests were supplied, so the init handshake and actual tool behavior cannot be verified. First-run timeout risk and a Python 3.13+ requirement add fragility, so the score is conservative.
Security and permissions 3/20: Red-line risk identified: PowerShell, Registry writes, FileSystem delete, and Process termination are enabled by default with no confirmation step, and remote transports (SSE/streamable HTTP) can expose them. Default telemetry also makes external network calls to PostHog. Mitigations exist (auth key, IP allowlist, TLS, OAuth, tool allowlisting), but the lack of confirmation for irreversible/remote commands places this in the 0-4 band.
Maintenance 10/20: The repository is not archived, has an MIT license, 10 open issues, and a published PyPI package/version (1.0.1), indicating active maintenance; however, no commit frequency, CI status, dependency-update cadence, or issue-response evidence was provided, leaving governance/versioning gaps.
Documentation 14/20: The README is layered: install/config examples for multiple clients, transport options, security configuration, environment variables, a tool list, telemetry disclosure, and limitations. Deductions: no detailed per-tool parameter reference, no dedicated troubleshooting section, and caveats such as the first-run timeout are mentioned but not deeply explained.
Setup experience 13/20: Setup is documented with uvx and copy-paste configs for Claude Desktop, Perplexity, Gemini, Qwen, Codex, and Claude Code, plus WSL and MSIX notes. Deductions: Python 3.13+ and uv are required, the first run may time out, and MSIX requires manual absolute-path configuration, which makes the process clear but somewhat fragile; no executable CI/test evidence caps the score.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkControls a browser

Best for

Developers and QA engineer needing UI automation on Windows
Individuals looking to integrate AI agents with desktop applications
System administrators requiring file and registry operations

Not for

Non-Windows platforms (e.g., macOS, Linux)
Scenarios requiring precise text selection or IDE programming support
Older Windows versions not covered by supported OS list

Required permissions

Access to UI elements via UI automation APIs
Execute PowerShell commands
Filesystem read/write and delete
Registry read/write
Access to clipboard
Start and manage processes

Risks and side effects

Has full system access and can perform irreversible operations (e.g., delete files, modify registry)
Remote access risk if exposed over network
UI automation may be unstable, especially under heavy load or complex UI
It may interfere with running applications or cause unexpected behavior

Setup

Before you start

Runtime：Python 3.13+ · Prebuilt binary

WINDOWS_MCP_AUTH_KEY optionalsecret Optional Bearer token for HTTP/SSE remote access; generate your own secret.

WINDOWS_MCP_OAUTH_CLIENT_SECRET optionalsecret Optional OAuth 2.0 client secret; pair with CLIENT_ID.

POSTHOG_API_KEY optionalsecret Optional override of the PostHog write key for telemetry; set empty to skip PostHog init.

Other optional settings (16)

WINDOWS_MCP_SCREENSHOT_SCALE optional Optional screenshot scale factor (0.1–1.0) to shrink images on high-res displays; nothing to obtain.

WINDOWS_MCP_SCREENSHOT_BACKEND optional Optional screenshot backend (auto/dxcam/mss/pillow); defaults to auto.

WINDOWS_MCP_PROFILE_SNAPSHOT optional Optional; set to 1 to emit per-stage timing logs for Screenshot/Snapshot.

WINDOWS_MCP_DISABLE_FLASH optional Optional; set to 1 to disable the orange flash border after screenshots.

WINDOWS_MCP_IP_ALLOWLIST optional Optional comma-separated allowed IPs/CIDR ranges for remote access.

WINDOWS_MCP_CORS_ORIGINS optional Optional comma-separated origins allowed for cross-origin browser requests.

WINDOWS_MCP_TOOLS optional Optional comma-separated whitelist enabling only the listed tools.

WINDOWS_MCP_EXCLUDE_TOOLS optional Optional comma-separated blacklist disabling listed tools (e.g. PowerShell,Registry).

WINDOWS_MCP_SSL_CERTFILE optional Optional TLS certificate path; pair with KEYFILE, can be self-signed via openssl.

WINDOWS_MCP_SSL_KEYFILE optional Optional TLS private key path; pair with CERTFILE.

WINDOWS_MCP_OAUTH_CLIENT_ID optional Optional OAuth 2.0 client ID for HTTP transports; must be pre-provisioned by you.

WINDOWS_MCP_STATELESS_HTTP optional Optional; set to 1 for session-less streamable-http to ease reconnects and scaling.

ANONYMIZED_TELEMETRY optional Optional; set to false to disable anonymous usage telemetry.

POSTHOG_HOST optional Optional PostHog host for telemetry; defaults to https://us.i.posthog.com.

WINDOWS_MCP_DEBUG optional Optional; set to 1 to enable DEBUG-level verbose logging.

WINDOWS_MCP_WATCHDOG optional Optional; set to off to disable the UIA focus watchdog for long-running stability.

Ensure Python 3.13+ and UV are installed (pip install uv).
For Claude Desktop, add the following to claude_desktop_config.json: {"mcpServers":{"windows-mcp":{"command":"uvx","args":["windows-mcp","serve"]}}}.
Fully restart Claude Desktop.
For other clients, refer to the README for specific steps.

claude_desktop_config.json

{
  "mcpServers": {
    "windows-mcp": {
      "command": "uvx",
      "args": [
        "windows-mcp",
        "serve"
      ]
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "windows-mcp": {
      "command": "uvx",
      "args": [
        "windows-mcp",
        "serve"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add windows-mcp -- uvx windows-mcp serve

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After starting the server (uvx windows-mcp serve), the client's tool list should show the 20 tools such as Screenshot, Click and Snapshot; calling Screenshot or using the PowerShell tool to query system info and getting a result confirms the connection works.

Troubleshooting

Ensure Python 3.13+ is used
Ensure 'English' is set as default language, else disable 'App-Tool'
First run may take 1-2 minutes to install dependencies
If encountering 'Connection closed' errors, use full path to uvx.exe
Check log files: `~/.windows-mcp/server.log` and `server.error.log`

Things to try

Once connected, you can ask your AI assistant things like:

Take a screenshot of my desktop and tell me which windows are open.
Open Notepad and type 'Hello, Windows-MCP' into it.
List the files and folders in C:\Users.
Show me the running processes and terminate the one using the most memory.

Tools 20

Click writes

Click on the screen at the given coordinates.

Type writes

Type text on an element, optionally clears existing text.

Scroll read-only

Scroll vertically or horizontally on the window or specific regions.

Move writes

Move mouse pointer or drag (with drag=True) to coordinates.

Shortcut writes

Press keyboard shortcuts like Ctrl+C, Alt+Tab.

Wait read-only

Pause for a defined duration.

WaitFor read-only

Wait until text, an active window, an element, or a focused element appears by polling UI state.

DisplayInventory read-only

Read display layout, work areas, effective DPI, and scale metadata.

Show 12 more tools

Screenshot read-only

Fast screenshot-first desktop capture with cursor position, active windows, and an image.

Snapshot read-only

Full desktop state capture for interactive element IDs, scrollable regions, or DOM-based browser extraction.

App writes

Launch an application by Start Menu name or executable path; resize, move, and switch between windows.

PowerShell destructive

Execute PowerShell commands.

FileSystem writes

Read, write, copy, move, delete, list, search, and inspect files and directories.

Scrape read-only

Scrape the entire webpage for information.

MultiSelect writes

Select multiple items (files, folders, checkboxes) with optional Ctrl key.

MultiEdit writes

Enter text into multiple input fields at specified coordinates.

Clipboard writes

Read or set Windows clipboard content.

Process destructive

List running processes or terminate them by PID or name.

Notification writes

Send a Windows toast notification with a title and message.

Registry destructive

Read, write, delete, or list Windows Registry values and keys.

Use cases

Automate UI interactions in desktop applications

File management and navigation

Execute PowerShell commands and scripts

Web scraping and browser automation

QA testing and desktop validation

Supported clients

Claude Desktop

Claude Code

Perplexity Desktop

Gemini CLI

Qwen Code

Codex

Autohand Code

Listed from the project's documentation, not tested by this site.

Overview

Windows-MCP is a lightweight, open-source MCP server that bridges the gap between AI agents and Windows OS, enabling agents to perform tasks such as file navigation, application control, UI interaction, QA testing, and more. It supports Windows 7 through 11. It provides a rich toolset for UI automation, including simulation of keyboard and mouse, capturing window/UI state, and accessing PowerShell, file system, registry, etc. The server is released under the MIT license.

Similar servers

Hevy MCP 77 · B

Manage your Hevy workout data from AI assistants via MCP.

★ 509 · Tools 26 Compare with this →

Firecrawl MCP Server 75 · B

Firecrawl's official MCP server for web search, scraping, and structured extraction for AI agents

★ 7.6k · Tools 11 Compare with this →

BioMCP 75 · B

One binary. One grammar. Evidence from the biomedical sources you already trust.

★ 654 · Tools 11 Compare with this →

Compartment 75 · B

Encrypted, fully offline long-term memory for AI agents

★ 579 · Tools 14 Compare with this →

Compare with other MCP servers

Source revision b74c507d9e70 Data synced 2026-10-12 Read the FMRS scoring method