SysControl MCP
Provides tools for managing Docker containers, images, and other Docker resources.
Provides tools for git operations, including status, diff, and file management.
Provides tools for managing Homebrew packages, including installation, updates, and querying.
Provides tools for interacting with iMessage, such as sending and reading messages.
Provides tools for performing Spotlight searches on the local system.
Click on "Install 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., "@SysControl MCPshow me my CPU usage and top processes"
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.
SysControl
An AI agent for your Mac or Windows 11 PC that answers questions about your system — and can extend itself with new tools on the fly.
92 real-time tools covering CPU, RAM, GPU, disk, network, processes, iMessage, email, clipboard, browser, weather, reminders, Docker, Time Machine, Wi-Fi, calendar, contacts, Notes, Homebrew, media control, file management, Spotlight search, spreadsheets, Word documents, PDFs, image generation, deep web research, sub-agent orchestration, code editing, git integration, and more. The agent picks the right tools automatically, runs them in parallel, and answers in plain English.
Three ways to run it — pick whichever fits your workflow:
How | Best for | |
App (macOS) | One-click native macOS (SwiftUI) experience — no setup required | |
App (Windows) | Native Windows 11 desktop app (Flet) — unzip and run | |
CLI |
| Terminal-first workflow, scripting, SSH sessions (macOS · Windows · Linux) |
Claude Desktop | MCP server | Using SysControl tools inside Claude Desktop |
All interfaces share the same agent, tools, and providers — they're interchangeable.
App (macOS)
A native SwiftUI app with streaming chat, Markdown rendering, chat history sidebar, and auto-save — no Python, no terminal, no dependencies to install. Just download, open, and configure your provider in Settings.
Download
Option A — Pre-built DMG (no Xcode required):
Download the latest SysControl.dmg from GitHub Releases, drag to Applications, then bypass Gatekeeper on first launch (the app is ad-hoc signed, not notarized):
xattr -r -d com.apple.quarantine /Applications/SysControl.appOr right-click the app → Open → Open the first time.
Option B — Install from source (compiles locally, bypasses Gatekeeper automatically):
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/ks6573/SysControl/master/swift/install.sh)"To update later: use Check for Updates in the app (⇧⌘U), or run syscontrol-update from Terminal.
To uninstall: re-run with --uninstall.
Option C — Build manually:
git clone https://github.com/ks6573/SysControl.git
cd SysControl/swift
./build.sh release
open .build/SysControl.appRequires: macOS 14 (Sonoma) or later. Options B and C also require Xcode Command Line Tools (
xcode-select --install).
Features
Streaming responses — tokens appear as they arrive with live Markdown rendering
Auto-save — every conversation is saved automatically with an LLM-generated title
Chat history sidebar — browse and delete past chats
Settings — switch between local (Ollama) and cloud providers in-app
In-app updates — check for new versions from the menu bar (⇧⌘U) or Settings; DMG users get a one-click download, source-install users update automatically
No setup — everything is configured through the app itself
First Launch
On first launch an onboarding sheet appears automatically:
Choose Local (Ollama) — requires Ollama running locally, no API key needed
Or choose Cloud — enter your API key
Click Done and start chatting
To change providers later, open Settings (⌘,).
Chat history is saved as Markdown in
~/.syscontrol/chat_history/— view, edit, or delete freely.
Related MCP server: Windows Operations MCP
App (Windows)
A native Windows 11 desktop app (built with Flet) with streaming chat, Markdown rendering, expandable tool-call cards, inline charts, a session sidebar, and in-app settings. It runs the same Python agent backend as the macOS app and CLI, bundled into a standalone .exe — no separate Python install to manage.
Download
Download the latest
SysControl-windows-x64.zipfrom GitHub ReleasesUnzip it anywhere (e.g.
%LOCALAPPDATA%\Programs\SysControl)Run
SysControl.exe
First launch: the build is not code-signed, so Windows SmartScreen may warn you. Click More info → Run anyway.
Requires: Windows 11 (x64). For local models, install Ollama for Windows and pull a model (
ollama pull qwen3:30b); for cloud, enter your Ollama Cloud API key during onboarding.
Build from source
git clone https://github.com/ks6573/SysControl.git
cd SysControl
uv sync --extra gui --extra build
uv run pyinstaller SysControl.spec # -> dist\SysControl\SysControl.exeOr run it directly without bundling: uv run syscontrol-gui.
Platform notes
The Windows app exposes the same agent and tools as macOS. System tools — CPU, RAM, GPU, disk, network, processes, battery, files, clipboard, screenshots, volume, Wi-Fi, system logs (Event Log), shell (PowerShell/cmd), web research, charts, spreadsheets/Word/PDF, sub-agents — work natively. macOS-only integrations (iMessage, Apple Mail, Calendar/Contacts, Notes, Time Machine, Homebrew) report that they're unavailable on Windows rather than failing.
CLI
A terminal agent that runs the same backend as the app. Two install paths:
Option A — One-line install (recommended):
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/ks6573/SysControl/master/install-cli.sh)"Installs uv if missing, then runs uv tool install against the GitHub repo, exposing syscontrol and syscontrol-server on your PATH in an isolated venv. No clone required.
To update later: run syscontrol --update (or /update inside the REPL); the standalone syscontrol-cli-update script is also still installed.
To uninstall: re-run with -- --uninstall.
Option B — From a clone (for development):
git clone https://github.com/ks6573/SysControl.git
cd SysControl
curl -LsSf https://astral.sh/uv/install.sh | sh # if uv not installed
uv sync
uv run agent.pyRequirements
macOS or Linux
Python 3.11+ (uv will fetch one if your system Python is older)
Ollama for local mode, or an Ollama Cloud API key for cloud mode
CLI Flags
syscontrol # interactive
syscontrol --provider local --model qwen3:30b # local, skip prompt
syscontrol --provider cloud --api-key sk-... # cloud, skip prompt (key is remembered)
syscontrol --provider cloud --no-save-key # cloud, prompt every time
syscontrol --continue # resume the most recent session
syscontrol --resume # pick a previous session from a list
syscontrol --coding --approval normal # coding agent, ask before edits/shell
syscontrol --coding --approval plan # read-only planning mode
syscontrol --coding --approval auto # auto-accept coding edits/shellCloned (Option B) users can substitute
uv run agent.pyforsyscontrolin any of the commands below.
The first time you enter your Ollama Cloud API key (via --api-key or the prompt), it's saved to ~/.syscontrol/cli_credentials.json (0600) so you won't be asked again. Use --no-save-key to opt out, or /logout from the REPL to forget it.
Coding Mode
The CLI can run as a coding agent with a narrowed code tool set: file search/read/edit, git status/diff, and shell commands. Approval modes:
Mode | Behavior |
| Read-only. The agent can inspect and produce an implementation plan, but edits and shell commands are blocked. |
| Reads are automatic; file writes and shell commands ask for approval in the terminal. |
| "Just do it" mode. Auto-accepts coding edits and shell commands for the session. |
Press Shift+Tab to toggle between system monitor mode and coding mode. Inside coding mode,
use /approval plan, /approval normal, or /approval auto to switch policies. The old
standard and nuke names still work as aliases.
Slash Commands & Keyboard Shortcuts
The interactive CLI feels like Codex / Claude Code: type / to pop a completion menu, @ to inline a file from the cwd, ! to run a one-off shell command, and the bottom toolbar shows the active model, provider, approval mode, and message count at all times.
Command | Description |
| Show all commands and keyboard shortcuts |
| Clear the screen |
| Clear conversation history (keeps system prompt) |
| List available tools, optionally filtered by substring |
| Show the active model and provider |
| Toggle system/coding mode or enter a coding sub-mode |
| Run a detected test command, or a provided command, in coding mode |
| Run a detected lint/typecheck command, or a provided command, in coding mode |
| Append a timestamped note to |
| Dump the full output of the most recent tool call |
| List recently saved CLI sessions |
| Generate a |
| Summarize the conversation; |
| Switch coding-mode approval policy |
| Check for and install the latest SysControl release |
| Forget the saved Ollama Cloud API key |
| Quit the session |
Key | Action |
| Submit a single-line buffer; insert a newline once the buffer is multi-line |
| Submit any non-empty buffer; exit on an empty buffer |
| Cancel the in-flight LLM/tool stream (press again within 1s to exit cleanly) |
| History navigation |
| Reverse history search |
| Complete the current slash command, |
| Toggle system/coding mode |
| Clear the screen |
| Always insert a newline (alternative to letting Enter add one in multi-line mode) |
Type @<partial> to pop a file picker scoped to the current directory (uses git ls-files when available; falls back to a recursive walk). Selected paths stay literal in the buffer and are inlined at submit time as fenced code blocks (capped at 64 KB per file).
Type !<command> to run a shell command directly — output prints inline and the LLM is bypassed. Requires allow_shell=true in ~/.syscontrol/config.json.
History is persisted to ~/.syscontrol/cli_history. Conversations auto-save to ~/.syscontrol/cli_sessions/; resume the most recent with syscontrol --continue or pick from a list with syscontrol --resume.
Local Mode (Ollama)
ollama pull qwen3:30b # recommended
ollama serve
syscontrol --provider localTool-calling capable models:
Model | Notes |
| Default. Best tool use and reasoning |
| Faster, lower memory — includes thinking mode |
| Lightweight alternative |
| Battle-tested fallback |
Models without native tool-calling (e.g.
gemma3) will error.
Cloud Mode (Ollama Cloud)
syscontrol --provider cloud
# Enter your key when prompted — not echoed or stored in shell historyGet a key at ollama.com/settings/keys. Default cloud model: gpt-oss:120b.
Ending a Session
Say any natural goodbye (bye, exit, quit, done, farewell, cya, goodnight, …) or press Ctrl-C. The agent will offer to save your session before exiting.
Session Memory
On exit you are prompted to save a short note about the session. Notes are appended to SysControl_Memory.md with a timestamp. On next startup, if the file exists its contents are injected into the system prompt so the agent has context from prior sessions. The file is append-only and plain text — edit or delete entries freely.
The agent can also save and recall memory mid-session via the read_memory and append_memory_note tools — no need to wait for exit.
Privacy: SysControl stores only what you explicitly save. Ollama processes queries locally by default.
Permissions & Security
Sensitive tools are disabled by default. Enable them in ~/.syscontrol/config.json:
{
"allow_shell": true,
"allow_messaging": true,
"allow_message_history": true,
"allow_screenshot": true,
"allow_file_read": true,
"allow_file_write": true,
"allow_calendar": true,
"allow_contacts": true,
"allow_accessibility": true,
"allow_tool_creation": true,
"allow_deep_research": true,
"allow_email": true,
"allow_notes": true,
"allow_brew": true,
"allow_agents": true
}Each disabled tool returns an error with the exact flag needed to enable it.
Claude Desktop Setup
1. Add the MCP server to your config
Platform | Config path |
macOS |
|
Windows |
|
{
"mcpServers": {
"system-monitor": {
"command": "/path/to/uv",
"args": ["run", "/absolute/path/to/SysControl/mcp/server.py"],
"env": {}
}
}
}Use which uv to get the uv path.
2. Set the system prompt — create a Claude Desktop Project and paste the contents of mcp/prompt.json into the Project Instructions field.
3. Restart Claude Desktop — system-monitor will appear in the MCP servers list.
Self-Extension
When you ask for something no tool covers, the agent offers to build it:
You: What song is playing in Spotify right now?
Agent: I don't have a tool for that. Want me to create one? (yes/no)
You: yes
Agent: ✓ Tool `get_spotify_track` installed. Restart and ask again.The agent writes a Python function, validates syntax, scans for dangerous patterns (eval, exec, etc.), and appends it to mcp/server.py. Requires:
{ "allow_tool_creation": true }Tools (92 total)
Monitoring
Tool | What it does |
| CPU load (total + per-core), clock frequency, inline bar chart |
| RAM and swap — used, available, percent, inline stacked chart |
| GPU load, VRAM, temperature per device (NVIDIA/pynvml), inline chart |
| Per-partition space and cumulative I/O counters |
| Cumulative bytes sent/received and per-interface status |
| Live disk read/write and network download/upload speed (MB/s) |
| Top N processes by CPU or memory |
| Single call: CPU + RAM + GPU + disk + network + top processes |
| Triage scan returning prioritized critical/warning alerts |
System & Hardware
Tool | What it does |
| Static profile: CPU model, core count, RAM, GPU VRAM, disks, OS |
| Percent, charging state, time remaining |
| CPU/motherboard sensors (Linux/Windows) |
| Boot time, uptime, 1/5/15-min load averages |
| Live pressure + specs + OC capability + upgrade feasibility + bottleneck analysis |
Process Management
Tool | What it does |
| Deep inspection of a PID: path, cmdline, user, RSS/VMS, threads, open files |
| Find processes by name (case-insensitive partial match) |
| SIGTERM (default) or SIGKILL a PID. Refuses critical system processes. |
Network & Connectivity
Tool | What it does |
| All active TCP/UDP connections with state and owning process |
| Pings gateway, Cloudflare, Google DNS in parallel and diagnoses slowness |
| Nearby networks with SSID, channel, security, signal strength |
Storage
Tool | What it does |
| Top N largest files under a path. Skips |
| Unmount and eject an external disk by mountpoint |
Messaging & Communication
Tool | What it does |
| Send an iMessage or SMS via Messages.app. macOS only. |
| Read recent messages from |
| Read recent emails from Mail.app (by folder). Requires |
| Send an email via Mail.app. Requires |
| Search emails across all accounts and mailboxes. Requires |
Browser & Web
Tool | What it does |
| DuckDuckGo search — title, URL, snippet. No API key. |
| Fetch a URL as plain text. No browser required. |
| Unlock browser control (called once, after user consent) |
| Open a URL in the default browser |
| Navigate the active tab via AppleScript (macOS) |
| Return the URL, title, and text of the current tab (macOS) |
Clipboard & Screen
Tool | What it does |
| Return current clipboard text |
| Write text to the clipboard |
| Full-screen PNG returned inline. Optionally save to file. macOS only. |
| Generate an inline visual image artifact from a prompt. Requires an OpenAI image API key. |
App Control & System
Tool | What it does |
| Open an app by name ( |
| Gracefully quit (AppleScript) or force-kill an app. macOS only. |
| Output, input, and alert volume; mute state |
| Set system output volume (0–100) |
| Currently playing track in Music.app or Spotify (title, artist, album, position). macOS only. |
| Play, pause, skip, or stop Music.app / Spotify. Auto-detects active player. macOS only. |
| Return the name of the focused application |
| Enable/disable Focus / DnD |
| Run a named Shortcut via |
File I/O & Shell
Tool | What it does |
| Read a text file (up to 16,000 chars) |
| Write text to any path, creating directories as needed |
| List directory contents with name, type, size, and modification time |
| Move or rename a file or directory |
| Copy a file to a new location |
| Delete a file or directory (Trash by default on macOS, recoverable) |
| Create a directory and any missing parents |
| Search for files system-wide using macOS Spotlight (mdfind). Instant. macOS only. |
| Read cells from |
| Write cells (A1 notation) or append rows to |
| Read paragraphs from |
| Find/replace text, overwrite paragraphs, or append to |
| Extract text from PDF files, page by page (up to 200 pages) |
| Execute a bash command and return stdout/stderr. Disabled by default. |
Calendar, Contacts & Logs
Tool | What it does |
| Upcoming events from Calendar.app for the next N days. macOS only. |
| Search Contacts.app by name — phone and email. macOS only. |
| List notes from Notes.app with title, folder, and timestamps. Requires |
| Read the full body of a note by title (partial match). Requires |
| Create a new note in Notes.app. Requires |
| Auto-start items (macOS LaunchAgents, Windows Registry, Linux |
| Last N lines of the system log with optional keyword filter |
Utilities
Tool | What it does |
| Schedule a macOS notification. Accepts |
| All pending reminders with IDs and fire times |
| Cancel a reminder by ID |
| Current weather + clothing recommendations. Auto-detects location from IP. |
| Homebrew, Mac App Store, and system software updates. macOS only. |
| List all installed Homebrew formulae and casks. Requires |
| Install a Homebrew formula or cask. Requires |
| Upgrade one or all Homebrew packages. Requires |
| Uninstall a Homebrew formula or cask. Requires |
| Running containers with live CPU%, memory, image, status, and ports |
| Last backup time, phase and progress if running, destination. macOS only. |
| Track UPS, USPS, FedEx, or DHL shipments by tracking number |
Memory
Tool | What it does |
| Read the persistent memory file — facts and notes saved across sessions |
| Append a concise note to the memory file for future recall |
Research
Tool | What it does |
| Multi-step web research agent: plans subquestions, searches multiple sources, extracts and cross-verifies claims, returns a citation-backed answer. Takes 1-3 minutes. |
Code Editing & Navigation
Tool | What it does |
| Read a file with line numbers, offset, and limit — large-file friendly. Requires |
| Targeted find-and-replace editing. Exact string match, fails if ambiguous. Requires |
| Find files by glob pattern (e.g. |
| Regex content search across files with optional context lines. Skips binary files. |
| Show branch, staged/unstaged/untracked files, and recent commits. |
| Show git diff for unstaged or staged changes. |
Sub-Agent Orchestration
Tool | What it does |
| List available sub-agents with names and descriptions |
| Delegate a focused task to a named sub-agent (explorer, analyst, researcher, writer) running in an isolated subprocess with restricted tools. Requires |
Self-Extension
Tool | What it does |
| Write, validate, and install a new MCP tool into |
| List all tools installed via |
Overclocking Support
Detected automatically from hardware and platform:
Platform | CPU OC | GPU OC |
Apple Silicon (M-series) | ✗ Not supported | ✗ Not supported |
Intel Mac | ✗ Not supported (no BIOS) | ✗ Not supported (macOS) |
Intel K/KF/KS — Windows/Linux | ✅ Intel XTU or BIOS | ✅ MSI Afterburner |
AMD Ryzen — Windows/Linux | ✅ Ryzen Master / PBO | ✅ MSI Afterburner |
Project Structure
SysControl/
├── agent.py # CLI entry-point shim
├── install-cli.sh # One-line CLI installer (curl one-liner)
├── agent/
│ ├── cli.py # Interactive terminal REPL
│ ├── core.py # Shared agent logic: MCP client, streaming loop, helpers
│ ├── bridge.py # JSON-over-stdio bridge for the Swift app
│ ├── agents.py # Sub-agent specs: AgentSpec, AgentRegistry, built-in agents
│ ├── runner.py # Sub-agent runner: isolated context, filtered tools
│ └── paths.py # Path resolution (repo root, user data dir, memory file)
├── mcp/
│ ├── server.py # MCP tool server (104 tools + self-extension)
│ └── prompt.json # System prompt for the agent
├── deep_research/ # Deep research agent (iterative web research with citation verification)
├── swift/
│ ├── Package.swift # SwiftPM package definition
│ ├── build.sh # Builds the .app bundle and DMG
│ ├── install.sh # One-line source installer
│ └── SysControl/ # SwiftUI source (App, Models, Views, Services, Storage)
├── scripts/
│ └── make_icon.py # Generates the .icns app icon from source PNGs
├── pyproject.toml # Python project config, dependencies, linting
├── VERSION # Current release version (single source of truth)
└── tests/ # Pytest suite for agent core + MCP helpersArchitecture
┌──────────────────────┐
│ SwiftUI App │ Native macOS frontend
│ (swift/SysControl/) │ Onboarding, chat, settings, history
└────────┬─────────────┘
│ JSON-over-stdio (bridge.py)
┌────────▼─────────────┐
│ Agent Core │ Streaming agentic loop, LLM client
│ (agent/core.py) │ Provider selection, tool dispatch
└────────┬─────────────┘
│ JSON-RPC (stdio)
┌────────▼─────────────┐
│ MCP Server │ 104 tools, self-extension, permission checks
│ (mcp/server.py) │ Concurrent tool execution via client pool
└──────────────────────┘The Swift frontend communicates with the Python backend through agent/bridge.py, which speaks a simple JSON-over-stdio protocol. The bridge reuses the same MCPClientPool and streaming loop used by the CLI, so all tools and capabilities are shared across every interface.
Entry Points
After installing via the one-liner (or pip install -e . from a clone):
Script | Description |
| Interactive CLI agent |
| Check for and install the latest release in place |
| Print the installed version and exit |
| MCP server (stdio) |
| Standalone updater script (same as |
From a clone without installing, use uv run:
Command | Description |
| Interactive CLI agent |
| MCP server (stdio) — for Claude Desktop integration |
License
This project is licensed under the MIT License. See LICENSE.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/ks6573/SysControl'
If you have feedback or need assistance with the MCP directory API, please join our Discord server