sysinternals-mcp
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., "@sysinternals-mcp list all processes with more than 100 threads"
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.
Sysinternals MCP
FastMCP 3.4 wrapper for 12 Sysinternals CLI tools (+ DIY RAMMap equivalent via WMI) -- autorunsc, handle64, pslist, listdlls, tcpvcon, sigcheck, accesschk, psloggedon, psfile, coreinfo, du, psinfo, rammap.
Stack: Python 3.12+ -- FastMCP 3.4.2 -- uv -- ruff
All 12 tools are confirmed CLI-native. No GUI automation, no fake backends.
First-run auto-downloads each binary from https://live.sysinternals.com/,
verifies the Authenticode signature, and caches it.
Tools
Tool | Function | What it does |
|
| Startup/persistence scan -- CSV output via |
|
| Open handles/file locks -- verbose via |
|
| Process list with CPU, thread, handle counts |
|
| Loaded DLLs per process with version info |
|
| TCP/UDP connections + owning process |
|
| Authenticode verification, version, VT lookup |
|
| Effective permissions on files/registry/services |
|
| Logged-on users, local + network |
|
| Remotely opened files on this machine |
|
| CPU topology, NUMA, cache, feature flags |
|
| Directory size breakdown (recursive) |
|
| Physical memory breakdown via WMI (DIY RAMMap, no binary) |
|
| System info: OS, uptime, hotfixes, services |
Related MCP server: Sysinternals MCP Server
Quick start
git clone https://github.com/sandraschi/sysinternals-mcp.git
cd sysinternals-mcp
uv sync --group dev
uv run sysinternals-mcp # stdio MCP for IDEFirst run auto-downloads binaries and stores them in %LOCALAPPDATA%\sysinternals-mcp\bin\.
Transport
Mode | Command |
stdio |
|
HTTP |
|
Binary handling
No EXEs committed to git. Binaries downloaded on first use from live.sysinternals.com.
Authenticode verification: rejects anything not signed by Microsoft / Sysinternals.
EULA: accepted once per machine (marker file in cache dir).
Excluded tools (GUI-only, no scriptable export)
RAMMap -- no CLI export flag exists.
Process Explorer -- GUI-only, no scriptable CLI output.
License
MIT
Available Tools
26 toolsaccesschkC
Check effective permissions on files, directories, registry keys, or services.
Wraps Sysinternals AccessChk.
Return Format
{"success": bool, "entries": list[dict], "count": int, "error": str | None}
Inputs: path (required), target, recursive, accept_eula. Returns: object
accesschk_serviceC
Check effective permissions on a specific Windows service.
Return Format
{"success": bool, "entries": list[dict], "count": int}
Inputs: service_name (required). Returns: object
autorunscC
Scan startup programs, scheduled tasks, services, drivers, and browser extensions.
Wraps Sysinternals Autorunsc with CSV output parsing.
Return Format
{"success": bool, "entries": list[dict], "count": int, "error": str | None}
Inputs: hashes, verify, all_users, accept_eula. Returns: object
autorunsc_scheduleB
Scan only scheduled tasks autoruns (fast subset).
Return Format
{"success": bool, "entries": list[dict], "count": int}
Inputs: hashes, accept_eula. Returns: object
coreinfoB
Show CPU topology, NUMA node layout, cache sizes, and feature flags.
Wraps Sysinternals Coreinfo.
Return Format
{"success": bool, "cpu_topology": list[dict], "features": list[dict], "error": str | None}
Inputs: all_features. Returns: object
duA
Show disk usage for a directory tree, broken down by subdirectory.
Wraps Sysinternals DU. Uses verbose (-v) output for per-directory breakdown.
Return Format
{"success": bool, "directories": list[dict], "total": dict, "error": str | None}
Inputs: path (required), n_children, accept_eula. Returns: object
du_quickA
Quick disk usage summary for a directory (one level, no recursion into children).
Return Format
{"success": bool, "directories": list[dict], "total": dict}
Inputs: path (required). Returns: object
handle64B
List open handles and file locks on the system.
Wraps Sysinternals Handle64.
Return Format
{"success": bool, "handles": list[dict], "count": int, "error": str | None}
Inputs: pattern, verbose. Returns: object
handle64_by_pidA
List open handles for a specific process by PID.
Return Format
{"success": bool, "handles": list[dict], "count": int}
Inputs: pid (required). Returns: object
listdllsB
List all loaded DLLs across all processes with version info.
Wraps Sysinternals ListDLLs.
Return Format
{"success": bool, "processes": list[dict], "count": int, "error": str | None}
Inputs: verified. Returns: object
listdlls_by_pidB
List loaded DLLs for a specific process by PID.
Return Format
{"success": bool, "dlls": list[dict], "count": int}
Inputs: pid (required). Returns: object
psfileB
List files opened by remote systems via network shares.
Wraps Sysinternals PsFile.
Return Format
{"success": bool, "files": list[dict], "count": int, "error": str | None}
Inputs: accept_eula, path_filter. Returns: object
psfile_closeC
Close a file opened by a remote system.
Return Format
{"success": bool, "message": str}
Inputs: file_id (required). Returns: object
psinfoB
Show detailed system information: OS version, uptime, hotfixes, services.
Wraps Sysinternals PsInfo.
Return Format
{"success": bool, "sections": dict, "raw": str, "error": str | None}
Inputs: show_hotfixes, show_services. Returns: object
psinfo_remoteB
Show system info for a remote computer.
Return Format
{"success": bool, "sections": dict, "raw": str}
Inputs: server (required). Returns: object
pslistB
List processes with PID, CPU time, thread count, and handle count.
Wraps Sysinternals Pslist.
Return Format
{"success": bool, "processes": list[dict], "count": int, "error": str | None}
Inputs: tree, include_kernel. Returns: object
pslist_detailC
List details for processes matching a name.
Return Format
{"success": bool, "processes": list[dict], "count": int}
Inputs: name (required). Returns: object
psloggedonC
List users logged on locally and optionally via network connections.
Wraps Sysinternals PsLoggedon.
Return Format
{"success": bool, "users": list[dict], "count": int, "error": str | None}
Inputs: local, network. Returns: object
psloggedon_serverB
List users logged on to a remote server.
Return Format
{"success": bool, "users": list[dict], "count": int}
Inputs: server (required). Returns: object
rammap_file_backedB
Show file-backed page cache summary from system working set.
Reports cached file extensions, top mapped files, and system cache totals. Uses WMI queries against the operating system cache manager.
Return Format
{"success": bool, "cache_total_mb": float, "sections": list[dict], "error": str | None}
Inputs: no named parameters. Returns: object
rammap_physicalA
Show physical memory usage breakdown by category.
RAMMap-style view: active, standby, modified, modified-no-write, transition, zeroed, free, and bad page counts. Also reports total, available, cached, and page-file sizes.
Uses a single PowerShell script that queries WMI and performance counters. No binary download required.
Return Format
{"success": bool, "categories": dict, "totals": dict, "error": str | None}
Inputs: detailed. Returns: object
rammap_processesB
Show per-process memory usage: working set, private bytes, shareable, pagefile.
RAMMap-style process list sorted by working set descending.
Uses Get-Process with WMI extensions for private working set data. No binary download required.
Return Format
{"success": bool, "processes": list[dict], "count": int, "total_ws_mb": float, "error": str | None}
Inputs: top_n. Returns: object
rammap_summaryA
Combined memory diagnostic: physical breakdown + top processes + file cache.
One-shot equivalent of calling rammap_physical, rammap_processes (top 10), and rammap_file_backed.
Return Format
{"success": bool, "physical": dict, "top_processes": list[dict], "file_cache": dict}
Inputs: no named parameters. Returns: object
rammap_treemapA
Generate a WizTree-style interactive treemap of process memory usage.
Creates a self-contained HTML file with an embedded D3.js squarified treemap, colored by process category (system, browser, dev, media, etc.) with hover tooltips showing PID, working set, and private bytes.
The HTML is fully self-contained (loads D3 from CDN) and opens in any browser.
Return Format
{"success": bool, "file_path": str, "processes": int, "total_mb": float}
Examples
rammap_treemap(top_n=100, min_ws_mb=0.5)Inputs: top_n, min_ws_mb. Returns: object
sigcheckB
Verify file digital signatures, version info, and optionally check VirusTotal.
Wraps Sysinternals Sigcheck with CSV output parsing.
Return Format
{"success": bool, "files": list[dict], "count": int, "error": str | None}
Inputs: path (required), recursive, accept_eula, virus_total. Returns: object
tcpvconA
List all TCP/UDP connections with owning process.
Wraps Sysinternals Tcpvcon with CSV output parsing.
Return Format
{"success": bool, "connections": list[dict], "count": int, "error": str | None}
Inputs: all_states. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 26 tools
Most tools target distinct Sysinternals utilities or have clear variant suffixes (local/remote, by_pid, quick, schedule), but some overlap exists (e.g., pslist vs pslist_detail, accesschk vs accesschk_service) where boundaries rely on subtle description cues.
All tool names use consistent snake_case with underscores for variants (e.g., _by_pid, _remote, _service, _quick), and there is no mixing of camelCase or other naming conventions.
With 26 tools, the set exceeds the typical 3-15 range; many are niche variants (e.g., five rammap_* tools, multiple duplicate base operations) making it heavy and potentially overwhelming for an agent to navigate.
The set covers a wide range of Windows diagnostics (processes, DLLs, handles, network, autoruns, memory, disk), but lacks action-oriented tools (e.g., process termination, service control) and some Sysinternals utilities (e.g., pskill, psloglist), leaving minor gaps.
Maintenance
Related MCP Connectors
Reasoning, code, anti-deception, memory harness MCP tools. Stdio or HTTPS api.ejentum.com/mcp
Connect ChatGPT and Claude to your Mac, Windows or Linux with secure MCP remote access.
Connect ChatGPT, Claude, Cursor or VS Code to your Windows PC: files, terminal, browser, apps.
Manage Superlist tasks and lists in plain language from any MCP-compatible AI agent.
Related MCP Servers
- AlicenseNot gradedqualityNot gradedmaintenanceEnables comprehensive Windows process and service management with secure UAC elevation. Supports discovery, inspection, and control of processes by name, PID, or port number, with both AI-assisted and direct command-line interfaces.-
- FlicenseNot gradedqualityDmaintenanceExposes Sysinternals and NirSoft Windows diagnostic binaries as MCP tools with safe subprocess execution. Dynamically registers tools from a binaries directory with built-in security filters for destructive operations.-
- FlicenseNot gradedqualityDmaintenanceA proof-of-concept MCP bridge that exposes Velociraptor's forensic triage tools to LLMs, enabling natural language querying of Windows endpoints for artifacts like network connections and suspicious processes.96-
- AlicenseNot gradedqualityCmaintenanceEnables AI assistants to perform comprehensive Windows system health diagnostics, security checks, developer workflows, and maintenance tasks via 85 tools with actionable recommendations, over MCP and OpenAI-compatible HTTP.MIT