system-informer-mcp
Give your AI agent full control of Windows. A native C MCP (Model Context Protocol) server built on System Informer / Process Hacker — 54 tools for processes, threads, memory, handles, services, network & more.
Install / Use
claude mcp add Xiaobocai08 -- npx -y github:Xiaobocai08/system-informer-mcpIf the server publishes to npm under a different name, use that package instead — check the repo README.
MCP Server
Model Context Protocol server
Quality Score
Category
SecuritySupported Platforms
Skill content
View source on GitHubSystem Informer MCP
Give your AI agent the full power of System Informer (Process Hacker) on Windows.
A lightweight, native Model Context Protocol server — written in C, ~350 KB, zero runtime dependencies — that lets an LLM inspect and control Windows the same way a human does through the System Informer GUI: processes, threads, memory, handles, services, network, drivers, windows and more.
⬇ Download the prebuilt si-mcp.exe (v1.0.0, x64) — no compiler needed
English | 简体中文
</div>TL;DR — Ask your AI: "What's using port 3000?", "Why is my CPU at 100%?", "Which process is locking this DLL?", "Is this .exe validly signed?", "Start Notepad, then suspend it." — and it just does it, through 54 well-documented native tools.
Why this exists
Modern System Informer builds have no headless / command-line mode — the switches only configure the GUI. So instead of scraping a GUI (fragile and lossy), this server does exactly what System Informer itself does: it compiles against System Informer's own phnt NT-API headers and calls the same native Windows APIs (NtQuerySystemInformation, NtQueryInformationProcess/Thread, the Service Control Manager, IP Helper, WinVerifyTrust, ...). Results come back as structured JSON instead of GUI tables.
That is the tightest practical coupling to how System Informer actually works — and it means your AI never has to guess.
Highlights
- 54 tools covering every System Informer data view and action.
- Native & tiny. Pure C, a single ~350 KB
si-mcp.exe, no Python/Node runtime, no DLLs to ship. - The AI never guesses. A rich
instructionsblock at startup, a fully documented JSON Schema for every argument (types, enums, defaults), and ready-made workflow prompts. - Safe by default. Every destructive/irreversible action is guarded behind an explicit
confirm: true; killing a critical process needs a second gate. Every handler runs under a structured-exception guard so a fault returns an error instead of crashing the server. - Honest about privileges.
server_statustells you exactly what you can and cannot see given your elevation.
See it in action
// Ask: "What is listening on port 3000?"
→ port_owner { "port": 3000 }
← { "port": 3000, "matches": 1,
"owners": [ { "protocol": "TCP", "local": "0.0.0.0:3000",
"state": "LISTEN", "pid": 18240,
"processImage": "C:\\Program Files\\nodejs\\node.exe" } ] }
// Ask: "Start Notepad, then suspend it."
→ launch_process { "command_line": "notepad.exe" } → { "pid": 51432 }
→ suspend_process { "pid": 51432, "confirm": true } → { "status": "ok" }
How it works
flowchart LR
A[AI Agent<br/>Claude / Cursor / ...] -- MCP over stdio --> B[si-mcp.exe<br/>~350 KB, C]
B -- phnt NT-API headers --> C[Native Windows APIs<br/>Nt* / SCM / IP Helper / WinTrust]
C --> D[Live System<br/>processes · threads · memory<br/>handles · services · network]
B -. hand-off .-> E[System Informer GUI<br/>live graphs · PE Viewer]
Quick start
1. Get the binary
Option A — download (recommended). Grab si-mcp-v1.0.0-windows-x64.exe from the latest release and put it anywhere. It is a single ~340 KB file with no dependencies. SHA256SUMS.txt is provided for verification.
Option B — build from source. Requires Visual Studio 2022 (or the free Build Tools) with the Desktop development with C++ workload and a Windows 10/11 SDK. Nothing else — no CMake, no package manager.
git clone https://github.com/Xiaobocai08/system-informer-mcp.git
cd system-informer-mcp
build.bat
:: -> build\si-mcp.exe
The phnt headers are vendored in third_party/phnt, so the repo builds on its own.
2. Register with your MCP client
<details open> <summary><b>Claude Code</b></summary>claude mcp add system-informer -- "C:\path\to\system-informer-mcp\build\si-mcp.exe"
</details>
<details>
<summary><b>Cursor / Claude Desktop / any stdio MCP client</b> (config JSON)</summary>
{
"mcpServers": {
"system-informer": {
"command": "C:\\path\\to\\system-informer-mcp\\build\\si-mcp.exe"
}
}
}
</details>
Run your client as Administrator for full coverage. Call
server_statusany time to see what you have.
3. Ask your AI
"Show me the top 5 processes by memory, then tell me if the biggest one is signed."
All 54 tools
| Group | Tools |
|---|---|
| System | server_status · system_overview · system_cpu_usage · system_memory · system_uptime |
| Processes | list_processes · process_tree · process_details · process_token |
| Process control | launch_process · launch_process_elevated · terminate_process · terminate_process_tree · suspend_process · resume_process · set_process_priority · set_process_affinity · empty_working_set · set_process_critical · create_process_dump |
| Threads | process_threads · thread_details · suspend_thread · resume_thread · terminate_thread · set_thread_priority · set_thread_affinity |
| Memory | process_memory_regions · process_memory_summary · read_process_memory · write_process_memory · protect_process_memory · search_process_memory · process_memory_strings |
| Handles | list_handles · handle_type_summary · find_handles_by_name · close_handle |
| Services | list_services · service_details · control_service · set_service_start_type · create_service · delete_service |
| Network | network_connections · port_owner |
| Modules / Drivers | process_modules · list_drivers |
| Windows | list_windows · window_action |
| Files | file_details · verify_file_signature |
| GUI hand-off | launch_systeminformer_gui · launch_peview |
Call tools/list for the full schema of each, and prompts/list for guided workflows: diagnose high CPU, triage a suspicious process, what-locks-this-file, what-owns-this-port, inspect a process.
Privileges
Windows — not this server — decides what is visible. Run elevated for full coverage:
| Running as | Coverage | |---|---| | Standard user | Your own processes in full; others limited; kernel addresses hidden | | Administrator | Nearly everything: all processes, service control, kernel addresses | | Admin + KSystemInformer driver | Protected processes (PPL, antimalware) too |
Safety model
Mutating or irreversible tools refuse to run without "confirm": true — terminate_*, suspend/resume, priority/affinity, empty_working_set, set_process_critical, write_process_memory, protect_process_memory, close_handle, all service mutations, and window_action. Terminating a Windows-critical process can bugcheck the machine, so that path has an extra allow_critical gate.
How it compares
| | system-informer-mcp | Typical "system info" MCPs |
|---|---|---|
| Language / size | Native C, ~350 KB exe | Python/Node + runtime |
| Depth | Processes, threads, memory r/w, handles, tokens, services, network, drivers, windows, signatures | Usually read-only process/CPU/mem stats |
| Control actions | Terminate, suspend, priority, affinity, dumps, service & handle control | Rarely |
| Guidance to the LLM | Rich instructions + per-arg JSON Schema + prompts | Minimal |
| Safety | confirm guard + SEH isolation | Varies |
Project layout
build.bat MSVC build script (no external build system)
src/
main.c stdio JSON-RPC loop + MCP dispatch
mcp.c / mcp.h tool registry, argument & result helpers
prompts.c startup instructions + workflow prompts
ntutil.c / ntutil.h NT/Win32 helpers (phnt): UTF-8, SIDs, errors
procsnap.c / procsnap.h shared process-snapshot helpers
tools_*.c one file per tool group
third_party/
cJSON/ vendored JSON library (MIT)
phnt/ vendored System Informer NT-API headers (MIT)
test/ Python MCP client + end-to-end tests
Testing
python test\mcp_client.py :: initialize + list every tool
python test\test_all.py :: exercise all 54 tools end-to-end
python test\roundtrip.py :: launch -> inspect -> control -> dump -> terminate
test_all.py launches a throwaway Notepad, drives all 54 tools against it and the live system, checks the confirm guards, and reports coverage. Latest run: 56/56 checks passed, 54/54 tools covered on Windows 11 (build 26200).
FAQ
Is this malware? No. It is an open-source, auditable MCP server. It exposes exactly what the System Informer GUI already exposes to any administrator, over a local stdio channel to your own AI client. Every dangerous action is gated behind confirm: true.
Does it need System Informer installed? No — every inspection/control tool works standalone. Only the two GUI hand-off tools (launch_systeminformer_gui, launch_peview) need the app; set SYSTEMINFORMER_PATH if it is installed in a non-default location.
macOS / Linux? No. This is deeply Windows-specific by design (NT native APIs).
Acknowledgements
- System Informer by Winsider Seminars & Solutions, Inc. — the
phntheaders and the tool this project mirrors. - cJSON by Dave Gamble.
- Model Context Protocol by Anthropic.
Star history
<a href="https://star-history.com/#Xiaobocai08/system-informer-mcp&Date"> <img src="https://api.star-history.com/svg?repos=Xiaobocai08/system-informer-mcp&type=Date" alt="Star History Chart" width="600"> </a>License
MIT. Builds against System Informer's phnt headers (MIT) and vendors cJSON (MIT).
<div align="center"> If this project helps you, please consider giving it a ⭐ — it genuinely helps others discover it. </div>
Related Skills
Agent-Reach
79.5kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
ruflo
72.1k🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
headroom
71.6kCompress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
CowAgent
46.9kOpen-source super AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-agent, multi-model, multi-channel. Lightweight, extensible, one-line install. (formerly chatgpt-on-wechat)
