WinHelm MCP
Uses headless Google Chrome (or Microsoft Edge) to convert Markdown and HTML documents into clean PDF files without requiring heavy dependencies like Puppeteer.
Provides NVIDIA GPU telemetry via nvidia-smi and WMI, reporting GPU status, memory, temperature, and utilization.
Allows binding the server to a Tailscale IP address for secure remote access across devices on a Tailscale network.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@WinHelm MCPwhat's my GPU temperature and usage?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
WinHelm — Windows Native MCP Server
WinHelm MCP — The helm for your Windows workspace.
A lightweight, production-grade Windows Native Model Context Protocol (MCP) server featuring a built-in Single-Process Web Gateway (Streamable HTTP/mcp+ Server-Sent Events/sse), Real-Time Web Monitor Dashboard, 38 System Tools with Dynamic Profile Loading, and MCP File Preview Resource.
📖 Tools Reference • ⚙️ Tool Profiles • 🛡️ Security Architecture • 💡 Agent Examples • 📝 Changelog • 🤝 Contributing
Table of Contents
Related MCP server: mcp-windows-server
Overview
WinHelm bridges the gap between AI assistants (Claude, Cursor, LibreChat, ChatGPT) and the Windows operating system. Unlike generic cross-platform servers, WinHelm is built from the ground up for Windows with:
Zero Gateway Overhead: Native single-process Web Gateway supporting both modern Streamable HTTP (
/mcp) and standard Server-Sent Events (/sse). No external proxy (supergatewayor reverse proxy) required.Safety First: Accidental deletions go to the Windows Recycle Bin instead of permanent destruction. Dangerous commands (
format-volume,rmdir /s /q c:\) are strictly blocked by default.First-class Windows Integrations: NVIDIA/WMI GPU telemetry, Windows Services management, native UTF-8 PowerShell runner, and headless Microsoft Edge PDF compilation.
Key Features
All-in-One Gateway & Web Monitor:
Streamable HTTP:
http://<HOST>:8788/mcp(Modern MCP standard)SSE Stream & POST:
http://<HOST>:8788/sseand/messageWeb Monitor Dashboard:
http://<HOST>:8788/(Interactive metrics, sessions, drives, memory, GPU, and live audit logs)Interactive File Viewer:
http://<HOST>:8788/preview?path=...(Rich Markdown & code viewer with syntax formatting)Health Check:
http://<HOST>:8788/health
High-Observability Logging & Secret Masking:
Microsecond execution timings with ANSI color-coded tags (
[HTTP],[TOOL-START],[TOOL-DONE],[SECURITY]).Automatically sanitizes Bearer tokens, API keys (
sk-***,ghp_***), AWS credentials (AKIA***), JSON properties ("authToken","password","x-api-key"), and sensitive arguments before printing or writing logs.
Background Daemon Task Engine:
Execute long-running dev servers (
npm run dev,docker compose up) as background tasks with detached PIDs, non-blocking logs streaming, and interactive stdin support.
High-Performance Code Search (Ripgrep):
Integrates
ripgrep(rg.exe) for fast regex search with streaming pagination (page,pageSize,hasMore) to prevent context window overflow.
Headless PDF Generation:
Converts Markdown and HTML into clean PDF documents using pre-installed Microsoft Edge or Chrome without requiring heavy dependencies like Puppeteer.
Tailscale & Remote Ready:
Optionally bind to
0.0.0.0or your Tailscale IP (100.x.y.z) for cross-device access. Default is127.0.0.1(loopback-only). Non-loopback binding requires Bearer Token verification and is guarded by a startup safety gate.
System Requirements
Component | Minimum Requirement | Recommended | Notes |
Operating System | Windows 10 / 11 / Server 2019+ | Windows 11 (64-bit) | Native Win32 & PowerShell APIs |
Node.js | Node.js >= 20.0.0 | Node.js 20+ LTS | Not required if using |
PowerShell | Windows PowerShell 5.1 | PowerShell 7+ (pwsh) | Auto-detects pwsh with UTF-8 encoding |
PDF Engine | Microsoft Edge | Pre-installed on Win 10/11 | Google Chrome is also auto-detected |
Search Engine | Built-in recursive search |
| Install via |
Privileges | Standard User | Standard User | Administrator is only needed for |
Architecture & Dashboard
System Architecture
flowchart TD
Client["AI Client (Claude / Cursor / Remote)"]
subgraph WinHelm["WinHelm Server (Port 8788)"]
subgraph Gateway["Single-Process Unified Gateway"]
MCP_HTTP["Streamable HTTP (/mcp)"]
MCP_SSE["SSE Gateway (/sse, /message)"]
Dashboard["Web Monitor (/ & /dashboard)"]
PreviewUI["File Preview (/preview)"]
Health["Health & Audit Export (/health, /api/monitor/*)"]
end
subgraph Security["Security & Governance"]
AuthGuard["Bearer Token Authentication"]
RateLimit["Rate Limiter (120 req/min)"]
PathGuard["Allowed Directories Guard"]
CmdGuard["Blocked Commands Guard"]
Sanitizer["Secret Masking (API Keys / Passwords)"]
end
subgraph Engines["Core Engine Layer"]
PSEngine["PowerShell UTF-8 Runner"]
TaskEngine["Background Daemon Task Engine"]
FSEngine["Filesystem & Safe Delete (Recycle Bin)"]
SearchEngine["Ripgrep Streaming Engine"]
PDFEngine["Edge / Chrome Headless PDF Engine"]
SysEngine["System, GPU & Windows Services Inspector"]
end
end
subgraph Windows["Windows Operating System"]
Win32["Win32 Shell / Recycle Bin"]
PowerShell["PowerShell CLI / Tasks"]
WMI_NVIDIA["WMI & nvidia-smi Telemetry"]
SCM["Windows Service Control Manager"]
end
Client <--> Gateway
Gateway --> Security
Security --> Engines
Engines <--> WindowsWeb Monitor & Live Dashboard
Access the built-in real-time dashboard in your browser at http://localhost:8788/:

Live Task Execution & Telemetry Streaming

Interactive File & Markdown Preview (preview://file)
Access syntax-highlighted code and rendered Markdown directly at http://localhost:8788/preview?path=<filepath>:

Quick Start
1. Installation & Setup
Option A: Standalone Executable (Recommended for Production)
Download the portable zero-dependency winhelm.exe from GitHub Releases or compile locally with npm run build:exe. Requires no Node.js runtime on Windows 10/11.
Option B: NPM CLI (Convenient for Dev & Testing)
# Install globally
npm install -g winhelm-mcp
# Run immediately
winhelm --profile dev
# Or run on-demand with npx (no global install required)
npx winhelm-mcp --profile devOption C: Clone & Build from Source
git clone https://github.com/dhammawatthumpra-coder/winhelm-mcp.git
cd winhelm-mcp
npm install
npm run build2. Start the Server
# Start standard server on port 8788 (full profile)
npm start
# Or start with a lightweight profile for coding agents (dev) or small models (minimal)
node dist/index.js --profile dev
# Or customize port and bearer token via CLI flags
node dist/index.js --port 8788 --auth my-secret-token
# Or run in Read-Only mode (disallows file edits, deletions, and killing processes)
node dist/index.js --read-only3. Verify Health Check
Visit http://localhost:8788/health in your browser. You should receive:
{
"status": "ok",
"server": "winhelm-mcp",
"version": "1.1.2",
"activeSessions": {
"sse": 0,
"streamableHttp": 0
},
"uptimeSeconds": 12,
"timestamp": "2026-09-28T08:00:00.000Z"
}Tool Profiles ("Load Only What You Need")
AI coding assistants perform much better when their context window isn't bloated with dozens of unneeded tool schemas. WinHelm implements a Dynamic Profile System so your agent sees only the tools it actually needs:
# 1. Minimal Profile: 6 essential tools for small models (Haiku, Llama 8B, local LLMs)
winhelm --profile minimal
# 2. Core Profile: 15 essential tools (terminal, file read/write/edit/search/hash, process list, telemetry)
winhelm --profile core
# 3. Developer Profile: 28 tools (Core + background tasks, ripgrep, zip archives, HTTP requests, PDF reports, system open & file preview)
winhelm --profile dev
# 4. SysAdmin Profile: 37 tools (All Core + tasks, ripgrep, archives, services, event logs, network, process kill, desktop automation & preview — all except pdf_generate)
winhelm --profile sysadmin
# 5. Full Suite: All 38 tools + interactive preview resource (default)
winhelm --profile fullProfile Inclusions
Profile | Active Tools | Key Inclusions | Context Window Savings |
| 6 |
| 🟢 ~85% token reduction |
| 15 |
| 🟢 ~60% token reduction |
| 28 | All Core + | 🟡 ~30% token reduction |
| 37 | All tools except | 🟠 Full Windows ops toolkit |
| 38 | All 38 tools + interactive HTML preview resource (default when omitted) | 🔵 Complete Windows control |
Token Economics & Model Optimization
Profile | Tools | Approx Context Tokens | Token Savings | Recommended Target Models & Use Cases |
| 6 | ~1,500 | 🟢 −85% | Claude 3.5 Haiku, Llama 3 8B, local LLMs or token-constrained pipelines |
| 15 | ~4,000 | 🟢 −60% | Everyday coding & file operations without background processes |
| 28 | ~6,800 | 🟡 −32% | Claude 3.7 Sonnet, GPT-4o, Cursor full-stack software development |
| 37 | ~9,200 | 🟠 −8% | Headless server management, Windows DevOps, diagnostics & audit |
| 38 | ~10,000 | 🔵 Baseline | Complete Windows native desktop suite with PDF & HTML visual previews |
Pro Tip: In
claude_desktop_config.json, pass["--profile", "dev"]underargsfor software development, or["--profile", "minimal"]for Claude Haiku!
Client Configuration
1. Claude Desktop (claude_desktop_config.json)
Path: %APPDATA%\Claude\claude_desktop_config.json
Option A: Zero-Install via npx (Easiest & Recommended)
{
"mcpServers": {
"winhelm": {
"command": "npx",
"args": ["-y", "winhelm-mcp", "--stdio", "--profile", "dev"],
"env": {
"MCP_ALLOWED_DIRECTORIES": "D:\\Workspace,C:\\Projects"
}
}
}
}Instant Setup: Runs the latest official
winhelm-mcpdirectly from npm on demand without manual cloning or global installation. Specify permitted workspace paths inMCP_ALLOWED_DIRECTORIES.
Option B: Remote / Local SSE Gateway
{
"mcpServers": {
"winhelm": {
"url": "http://127.0.0.1:8788/sse",
"headers": {
"Authorization": "Bearer my-secret-token"
}
}
}
}Option C: Direct Stdio / Node Process (From Source)
{
"mcpServers": {
"winhelm": {
"command": "node",
"args": ["<PATH_TO_WINHELM>/dist/index.js", "--stdio", "--profile", "dev"],
"env": {
"MCP_ALLOWED_DIRECTORIES": "D:\\mcp,C:\\Projects"
}
}
}
}Context Optimization: Supplying
"--profile", "dev"restricts tools to 28 developer essentials, saving ~32% context tokens while preserving all coding, ripgrep, background task, PDF, archive, and preview capabilities.
Option D: Standalone Executable (winhelm.exe)
{
"mcpServers": {
"winhelm": {
"command": "<PATH_TO_WINHELM>\\dist\\winhelm.exe",
"args": ["--stdio", "--profile", "dev"]
}
}
}Option E: Per-Project Dedicated Config File (Isolated Profiles & Roots)
{
"mcpServers": {
"winhelm-project-a": {
"command": "winhelm",
"args": ["--stdio", "--config", "D:\\mcp\\configs\\project-a.json"]
}
}
}Multi-Instance Isolation: Each project config file maintains its own isolated
allowedDirectories,profile, and security rules without modifying the shared defaultwinhelm.config.json. CLI overrides are session-only (--no-persistby default) to prevent instances from colliding.
2. Cursor IDE
In Cursor Settings (Settings -> Features -> MCP -> Add new MCP server):
Name:
winhelmType:
sseURL:
http://127.0.0.1:8788/sse
(If authentication is configured, add "Authorization": "Bearer <YOUR_TOKEN>" to the headers section).
3. ChatGPT & OpenAI MCP Tunnel (tunnel-client)
WinHelm integrates seamlessly with OpenAI's official tunnel-client daemon, allowing ChatGPT to execute Windows commands, inspect files, and manage background tasks directly over a secure Cloudflare Tunnel:
Profile Configuration (winhelm-mcp.yaml)
config_version: 1
control_plane:
base_url: "https://api.openai.com"
tunnel_id: "your-tunnel-id-here"
api_key: "env:CONTROL_PLANE_API_KEY"
health:
listen_addr: "127.0.0.1:18026"
admin_ui:
open_browser: false
log:
level: info
format: json
mcp:
commands:
# Direct stdio connection with dev profile (~32% token savings for ChatGPT)
- channel: main
command: 'node D:/mcp/winhelm-mcp/dist/index.js --stdio --profile dev'Running the Tunnel
.\tunnel-client.exe run --profile-file winhelm-mcp.yamlPurity Guard: In
--stdiomode, WinHelm routes all operational logs tostderr, leavingstdoutpurely for JSON-RPC messages to guarantee zero parsing errors on ChatGPT.
4. Remote & Tailscale Connection
WinHelm binds by default to 127.0.0.1 (localhost only). To allow secure cross-device access over private networks like Tailscale or WireGuard, bind to 0.0.0.0 or your Tailscale IP:
Retrieve your machine's Tailscale IP (e.g.
100.80.20.10) or Tailscale Funnel domain (e.g.https://your-node.ts.net).Start WinHelm with
--host 0.0.0.0and a strong authentication token:node dist/index.js --port 8788 --host 0.0.0.0 --auth super-secure-token-here⚠️ Host Safety Gate: Binding to
--host 0.0.0.0exposes the server to your local network and Tailscale. You must supply an authentication token (--auth), otherwise startup will be rejected with exit code 1 by the host safety gate.Connect your mobile or remote Claude / Cursor / ChatGPT client:
Streamable HTTP:
http://100.80.20.10:8788/mcpSSE Stream:
http://100.80.20.10:8788/sseWeb Monitor Dashboard:
http://100.80.20.10:8788/?token=super-secure-token-here(or Tailscale Funnel URL)Header:
Authorization: Bearer super-secure-token-here
Claude.ai Custom Connectors (URL Query Token)
Claude.ai Custom Connectors and certain web/mobile clients do not provide a UI field to enter custom HTTP headers (such as Authorization: Bearer <token>). WinHelm natively supports passing the authentication token directly via the URL query parameter:
Server URL:
https://<your-tailnet-domain>.ts.net/mcp?token=<YOUR_AUTH_TOKEN>Authentication: Select
No sign-inTransport (under Advanced): Streamable HTTP (Default)
🔒 Security Guarantee: Passing the token in the URL query parameter still triggers full timing-safe cryptographic verification on the server, ensuring your Windows machine remains completely protected from unauthorized internet access without needing a complex OAuth setup.
Tools Overview (38 Tools)
WinHelm provides 38 focused Windows native tools grouped across 6 functional categories. Each tool is loaded dynamically based on your active --profile:
Category | Tools | In Profiles | Summary |
Terminal & Background Tasks | 6 |
| Synchronous PowerShell runner and detached daemon processes with live logs & stdin. |
Filesystem, Safe Delete & Archives | 12 |
| Surgical file edits, streaming tails, SHA-256 hashes, .NET zip archives, and Recycle Bin safe delete. |
Codebase Search, PDF & Preview | 3 |
| Streaming paginated |
Desktop, Clipboard & Toast | 5 |
| Windows clipboard read/write, primary screen capture, system app launcher ( |
System, Processes & Services | 9 |
| CPU/RAM/Drive telemetry, NVIDIA GPU stats, process list/kill, port inspector, event logs, and service control. |
Network & Connectivity | 3 |
| HTTP latency probe, full REST client ( |
📖 See docs/TOOLS.md for full parameter specifications, types, returns, and schemas.
MCP Resources & Web Endpoints
MCP Resources
WinHelm exposes 1 dedicated Model Context Protocol resource:
URI | MIME Type | Description |
|
| Dynamically renders rich HTML preview for Markdown and source code files with line numbers and syntax highlighting. |
Gateway Web Endpoints
WinHelm hosts a full web application on a single port (default: 8788):
/&/dashboard— Real-time Web Monitor Dashboard (CPU, RAM, GPU, sessions, requests, live logs)./mcp— Streamable HTTP transport endpoint./sse— Server-Sent Events transport endpoint for clients like Claude Desktop and Cursor./message— POST endpoint for incoming JSON-RPC messages in SSE mode./preview— Interactive browser file viewer (/preview?path=D:\project\README.md)./health— JSON health status and server uptime probe./api/monitor/stats— JSON hardware and session statistics./api/monitor/export?format=json|csv— Audit log export for security compliance.
💡 Browser Dashboard Access with Authentication:
When authentication is enabled (--auth <token>orauthToken), accessing the Web Monitor Dashboard or Previewer via your browser requires passing the token once in the URL:
Dashboard:
http://<HOST>:8788/?token=<YOUR_AUTH_TOKEN>(orhttps://<your-tailnet-domain>.ts.net/?token=<YOUR_AUTH_TOKEN>)File Preview:
http://<HOST>:8788/preview?path=D:\project\README.md&token=<YOUR_AUTH_TOKEN>
WinHelm validates the token, displays live metrics and real-time logs, and sets a secureHttpOnlysession cookie so subsequent dashboard navigation stays authenticated without re-entering the token.
Configuration & Environment Variables
WinHelm loads configuration in the following order of precedence:
CLI Flags
Environment Variables
winhelm.config.jsonDefault settings
Reference Table
CLI Flag | Environment Variable | Default | Description |
| N/A | auto | Load configuration from a specific JSON file (for per-project multi-agent isolation). |
| N/A |
| Run in standard I/O mode for local MCP clients (OpenAI tunnel-client, Claude, Cursor). |
|
|
| Tool profile to load: |
| N/A | auto | Explicit comma-separated tools to load or |
|
|
| Transport mode: |
|
|
| Port number for the Web Gateway and MCP server. |
|
|
| Network interface to bind ( |
|
| none | Bearer token for authentication. Rejects unauthenticated requests with HTTP 401. |
|
|
| Enables read-only mode (strictly blocks file writing, shell execution, process killing, service modification, and mutating HTTP requests). |
|
|
| Comma-separated directory paths permitted for file access (e.g. |
| N/A |
| Keep CLI overrides session-only without writing to |
| N/A |
| Persist CLI overrides back to the active configuration file. |
N/A |
| (see below) | Additional comma-separated commands to block from execution. |
N/A |
|
| Comma-separated allowed hostnames for Host header validation (DNS Rebinding protection). |
winhelm.config.json
Create or modify winhelm.config.json in your project root or %USERPROFILE%\.winhelm\config.json:
{
"blockedCommands": [
"format-volume",
"format-disk",
"clear-disk",
"stop-computer",
"restart-computer",
"rmdir /s /q c:\\",
"del /f /s /q c:\\"
],
"allowedDirectories": ["D:\\mcp", "C:\\Workspace"],
"allowSystemExecution": true,
"profile": "dev",
"fileReadLineLimit": 2000,
"defaultTimeoutMs": 60000,
"telemetryEnabled": false,
"authToken": null,
"readOnly": false,
"rateLimitWindowMs": 60000,
"rateLimitMaxRequests": 120
}Security Boundaries vs. Defense-in-Depth:
True Security Boundaries: Authentication Token (
authToken), Tool Profiles (minimal,core), and Read-Only Mode (readOnly). For untrusted or public environments, configure anauthTokenand use--profile coreto exclude terminal execution completely.Defense-in-Depth: The command blocklist protects against accidental destructive commands (
format-volume,rmdir /s /q c:\), but is not an impenetrable sandbox.allowedDirectoriesstrictly bounds native File Tools (file_read,file_write,create_zip, etc.) and default CWD.DNS Rebinding Defense: WinHelm validates
Hostheaders against an allowlist (localhost,127.0.0.1,[::1],*.ts.net). Requests from unauthorized hostnames are rejected with HTTP 403 Forbidden.Remote Tunnel Fail-Closed Gate: Requests arriving through a reverse proxy or tunnel without an
authTokenconfigured are rejected immediately with HTTP 403 Forbidden.Fail-Closed Filesystem:
allowedDirectories: []blocks all filesystem operations by default. Specify target paths (e.g.["D:\\mcp", "C:\\Workspace"]). Wildcard["*"]is strictly discouraged for unauthenticated or public network exposures.
Windows Background Service (Always-On Daemon)
WinHelm includes a built-in service manager using Windows Task Scheduler to run reliably in the background across system reboots:
# Install as a persistent Windows Service (Run PowerShell as Administrator)
npm run service:install
# Check service status
npm run service:status
# Stop / Start the service
npm run service:stop
npm run service:start
# Uninstall the service
npm run service:uninstallStandalone Single-File Executable (winhelm.exe)
You can compile WinHelm into a standalone .exe (~2.6 MB) that requires no Node.js installation on target machines:
npm run build:exeThe resulting executable will be generated at dist/winhelm.exe:
# Run standalone executable with custom port
.\dist\winhelm.exe --port 9000 --auth my-tokenSecurity, Rate Limiting & Auditing
Automatic Secret Redaction:
Automatically sanitizes sensitive keys (
sk-...,ghp_...,Bearer ********, and password values) from terminal output, dashboard UI, and log files.
Built-in Rate Limiting (Sliding Window):
Enforces a sliding window ceiling of 120 requests per minute per IP address, preventing runaway client loops.
Auditing & Log Rotation:
Logs are stored in
logs/winhelm-YYYY-MM-DD.log(capped at 10 MB per file, auto-pruning logs older than 7 days).Export audit logs anytime via browser or API:
http://localhost:8788/api/monitor/export?format=jsonhttp://localhost:8788/api/monitor/export?format=csv
Fail-Closed Filesystem Confinement:
allowedDirectories: []blocks all filesystem operations by default for safety.Symlinks and NTFS directory junctions are resolved via
fs.realpathSyncbefore boundary evaluation.
Loopback-First Network Binding:
Default host is
127.0.0.1(loopback only).Non-loopback binding (such as
0.0.0.0) without--authis blocked at startup with exit code 1.Bearer tokens are compared using
crypto.timingSafeEqualwith buffer length validation to prevent timing side-channels.
Troubleshooting
1. Port Conflict (EADDRINUSE: address already in use :::8788)
Cause: Another process or previous instance is using port 8788.
Solution: Specify a different port using
--port:node dist/index.js --port 8790Or check which process is holding port 8788:
Get-NetTCPConnection -LocalPort 8788 | Select-Object OwningProcess
2. HTTP 401 Unauthorized
Cause: WinHelm was started with
--auth <token>orMCP_AUTH_TOKEN, but the MCP client didn't supply matching credentials.Solution 1 (Clients with Custom Header Support): Add the Bearer token header to your client configuration:
"headers": { "Authorization": "Bearer <YOUR_TOKEN>" }Solution 2 (Claude.ai Custom Connectors without Header UI): Append the token directly to the Server URL and select No sign-in:
https://<your-domain>/mcp?token=<YOUR_TOKEN>Solution 3 (Browser Dashboard & File Preview): Append
?token=<YOUR_TOKEN>to the URL in your browser:https://<your-domain>/?token=<YOUR_TOKEN> https://<your-domain>/preview?path=...&token=<YOUR_TOKEN>
3. PowerShell Execution Policy Restriction
Cause: Windows blocks script execution (
File cannot be loaded because running scripts is disabled on this system).Solution: Run with bypass flag:
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
4. Windows Service Installation Error (Access is denied)
Cause:
npm run service:installregisters a task in Windows Task Scheduler, which requires elevated privileges.Solution: Open PowerShell as Administrator and re-run
npm run service:install.
5. Tailscale / Remote Timeout
Cause: Windows Defender Firewall is blocking inbound connections on port 8788.
Solution: Allow port 8788 in Windows Firewall:
New-NetFirewallRule -DisplayName "WinHelm MCP Gateway" -Direction Inbound -LocalPort 8788 -Protocol TCP -Action Allow
6. PDF Generation Headless Browser Not Found
Cause: Neither Microsoft Edge nor Google Chrome could be located in default system paths.
Solution: Ensure Microsoft Edge is installed at its standard location (
C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe) or install Google Chrome.
7. Host Safety Gate Blocks Startup (Non-loopback binding requires --auth)
Cause: You attempted to bind to a non-loopback address (such as
0.0.0.0) without specifying an authentication token.Solution: Add
--auth <token>or set theMCP_AUTH_TOKENenvironment variable. This security gate prevents accidental public exposure of your host system without authentication.
Documentation & Deep Dive
For in-depth guides, architectural references, and developer guidelines, explore the docs/ folder:
📖 Comprehensive Tools Reference: Exhaustive documentation for all 38 tools, including parameter types, options, return formats, and JSON-RPC examples.
⚙️ Tool Profiles & Context Optimization: Deep dive into the 5 built-in profiles (minimal, core, dev, sysadmin, full), custom
--toolsfiltering, token economics, and LLM optimization recipes.🛡️ Security Model & Architecture: Deep dive into the 5-layer security model, path confinement, regex command blacklists, and secret masking.
💡 Real-World Agent Examples: End-to-end workflows showing how AI agents build projects, troubleshoot Windows crashes, and generate executive PDFs.
📝 Changelog: Release notes and version history following Keep a Changelog.
🤝 Contributing Guide: Instructions for developing, running tests, and opening Pull Requests.
License
This project is licensed under the MIT License — see the LICENSE file for details.
Built cleanly from the ground up for the Windows Model Context Protocol developer community.
This server cannot be deployed
Maintenance
Related MCP Connectors
- mcp-serverOAuthcom.make
Give your AI agents the tools to build, manage, and run automation workflows.
Governed memory and workspace for any AI: tasks, calendar, mail and pages, with per-action consent.
Git-backed platform for skills, tools, and context for AI agents
Hosted runtime for persistent agent teams, durable workflows, memory, schedules, and goals.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to interact with Windows operating systems by providing tools for UI automation, file navigation, application control, and system operations. Works with any LLM to perform tasks like clicking, typing, launching applications, and executing PowerShell commands through native Windows integration.MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to control Windows systems through natural language commands, providing 200+ automation tools for system control, file operations, web automation, and more.56MIT
- AlicenseNot gradedqualityAmaintenanceEnables AI clients to securely control and interact with a local Windows machine through 218 configurable tools for files, Git, processes, Windows UI, browser automation, WSL, Office, recovery, skills, and child MCP servers.11 npmMIT
- AlicenseNot gradedqualityBmaintenanceProvides a Windows-first local AI-agent gateway with configurable tools for files, Git, processes, Windows automation, WSL, browser control, durable agent runs, memory, verification, and intelligent routing, while exposing a secure MCP endpoint for ChatGPT Web and a local web UI.10 npmMIT