wintools-mcp
Valhuntir Windows forensic tool execution via MCP
Install / Use
claude mcp add AppliedIR -- npx -y github:AppliedIR/wintools-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
Development & EngineeringSupported Platforms
Tags
Our assessment of wintools-mcp
wintools-mcp scores 78/100 on our quality scale, 3690th of 4,700 Development & Engineering skills we index.
Its MCP Server is 20 KB long, well organised into 29 sections with 11 code examples: a thorough specification that gives an agent plenty to work with.
It has 10 GitHub stars, so there is little community track record yet; judge it on its content.
Maintenance, license and trust
- The repository was last updated today, so wintools-mcp is actively maintained.
- Our last check on 2026-09-16 found the source still online.
- It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-11-13. Automated pattern scan on 2026-11-13. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
wintools-mcp compared with similar skills
All 4 of these similar skills score higher than wintools-mcp; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| wintools-mcp (this skill)by AppliedIR | 78 | 10 | today | MCP Server |
| taste-skillby Leonxlnx | 100 | 116.8k | 8d ago | CLAUDE.md |
| Agent-Reachby Panniantong | 100 | 112.3k | 9d ago | CLAUDE.md |
| rufloby ruvnet | 100 | 81.9k | today | CLAUDE.md |
| headroomby headroomlabs-ai | 100 | 79.8k | today | CLAUDE.md |
Frequently asked questions
- How do I install wintools-mcp?
- Run
claude mcp add AppliedIR -- npx -y github:AppliedIR/wintools-mcp. The install tabs above show the steps for each supported agent. - Which AI agents does wintools-mcp work with?
- It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
- Is wintools-mcp safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It is MIT-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is wintools-mcp still maintained?
- The repository was last updated today, so wintools-mcp is actively maintained.
Skill content
View source on GitHub
Windows Tools MCP
Catalog-gated Windows forensic tool execution with knowledge-enriched response envelopes.
Platform Documentation · Deployment Guide
Important Note — While extensively tested, this is a new platform. ALWAYS verify results and guide the investigative process. If you just tell Valhuntir to "Find Evil" it will more than likely hallucinate rather than provide meaningful results. The AI can accelerate, but the human must guide it and review all decisions.
Architecture
wintools-mcp runs independently on a Windows forensic workstation, exposing a Streamable HTTP endpoint on port 4624. The LLM client and vhir CLI are the two human-facing tools. The vhir CLI always runs on the SIFT workstation — it requires direct filesystem access to the case directory. When the LLM client runs on a separate machine, the examiner must have SSH access to SIFT for all CLI operations. The LLM client connects to wintools-mcp over the network.
graph LR
subgraph sift ["SIFT Workstation"]
CC["LLM Client<br/>(human interface)"]
CLI["vhir CLI<br/>(human interface)"]
CASE["Case Directory"]
CLI --> CASE
end
subgraph winbox ["Windows Forensic Workstation"]
WAPI["wintools-mcp API<br/>:4624"]
WM["wintools-mcp<br/>Windows tool execution"]
FK["forensic-knowledge"]
WAPI --> WM
WM --> FK
end
CC -->|"streamable-http"| WAPI
WM -->|"SMB"| CASE
Execution Pipeline
Every tool execution flows through the same security and enrichment pipeline.
graph LR
REQ["MCP tool call"] --> DENY{"Hardcoded<br/>Denylist"}
DENY -->|"blocked binary"| BLOCK["BLOCKED"]
DENY -->|"pass"| CAT{"YAML Catalog<br/>Allowlist"}
CAT -->|"not cataloged"| REJECT["REJECTED"]
CAT -->|"cataloged"| SANITIZE{"Argument<br/>Sanitization"}
SANITIZE -->|"dangerous flag<br/>or metachar"| REJECT
SANITIZE -->|"clean"| RESOLVE["Resolve<br/>Binary Path"]
RESOLVE --> EXEC["subprocess.run<br/>shell=False"]
EXEC --> PARSE["Parse Output<br/>CSV / JSON / text"]
PARSE --> ENRICH["FK Enrichment<br/>caveats, corroboration,<br/>field notes, discipline"]
ENRICH --> RESP["Response Envelope"]
Security Model
This MCP opens attack vectors where connected LLM clients can execute tools on this system. It must only be installed on dedicated forensic workstations that are isolated behind firewalls on a trusted network segment. Never install on personal laptops, production systems, or machines containing data outside the scope of the investigation.
The installer requires typing security_hole or passing -AcknowledgeSecurityHole before proceeding. This is an intentional friction point.
The additional controls in place (catalog allowlists, denylist of dangerous binaries, argument sanitization, shell=False execution) are defense-in-depth measures, not preventative controls. The MCP is not hardened and should never be deployed facing the Internet or other untrusted systems.
graph TB
INPUT["Tool Execution Request"] --> D{"Hardcoded Denylist<br/>20 binaries + .exe variants"}
D -->|"cmd, powershell, pwsh,<br/>wscript, cscript, mshta,<br/>rundll32, regsvr32, certutil,<br/>bitsadmin, msiexec, bash,<br/>wsl, sh, msbuild, installutil,<br/>regasm, regsvcs, cmstp, control"| BLOCKED["BLOCKED"]
D -->|"pass"| A{"YAML Catalog<br/>Allowlist"}
A -->|"unknown binary"| REJECTED["REJECTED"]
A -->|"cataloged"| S{"Argument<br/>Sanitization"}
S -->|"shell metacharacters,<br/>execution flags, or<br/>@file response-file"| REJECTED
S -->|"clean"| E["EXECUTE<br/>subprocess.run(shell=False)"]
All execution uses subprocess.run(shell=False). Only tools defined in YAML catalog files can run. Dangerous binaries (cmd, powershell, wscript, etc.) are unconditionally blocked by a hardcoded denylist. Arguments are checked for shell metacharacters, dangerous flags (execution-related: --exec, --command, -enc), and response-file syntax (@filename).
What wintools won't run
wintools runs tools on data already in the case. It doesn't collect: a command is refused, before any path in it is touched, when it
- names another system: a
\\host\...path (also//host/...,\\?\UNC\host\...and\??\UNC\host\...) whose host isn't the case share's host, or a device (\\.\..., a volume path); - puts a tool in its own live mode on the Windows host: sigcheck
-v,-o,-t; autorunsc-v, or without-z <systemroot> <userprofile>inside the case; Hayabusa--live-analysis/-landupdate-rules; EvtxECmd, SQLECmd and RECmd--sync; SBECmd-l; AppCompatCacheParser without-f <SYSTEM hive>; yara with a process id as its target; KAPE target mode (--tsource,--target,--tdest),--sync,--rlcand the upload switches (--sc*,--s3*,--as*,--sftp*).
Not checked: a KAPE module that itself collects live data, a _kape.cli file of saved KAPE commands, and a file a tool reads (a rules file).
Quick Start
On the Windows forensic workstation, clone the repo:
git clone https://github.com/AppliedIR/wintools-mcp.git; cd wintools-mcp
Or if git is not installed, download and extract the ZIP:
Invoke-WebRequest https://github.com/AppliedIR/wintools-mcp/archive/refs/heads/main.zip -OutFile wintools.zip
Expand-Archive wintools.zip -DestinationPath . -Force; cd wintools-mcp-main
Then run the installer. Windows 10/11 refuses to run scripts by default, so the first line allows them for this PowerShell window only:
Set-ExecutionPolicy -Scope Process Bypass -Force
.\scripts\setup-windows.ps1
See SETUP.md for detailed deployment options and SMB configuration.
MCP Tools (10 total)
Discovery (6 tools)
| Tool | Description |
|------|-------------|
| scan_tools | Scan for all cataloged forensic tools, report availability and install guidance |
| list_windows_tools | List all cataloged tools with installation status, filterable by category |
| list_missing_windows_tools | List tools not installed, with installation guidance and alternatives |
| check_windows_tools | Check specific tools by name for availability |
| get_windows_tool_help | Get tool-specific help, flags, caveats, and interpretation guidance |
| suggest_windows_tools | Given an artifact type, suggest relevant tools and check availability |
Evidence Access (1 tool)
| Tool | Description |
|------|-------------|
| get_share_info | Get SMB share paths for evidence access (share_root, case_dir, evidence_dir, extractions_dir) |
KAPE Discovery (1 tool)
| Tool | Description |
|------|-------------|
| list_kape_targets | List the installed KAPE modules in structured categories. wintools runs KAPE modules over files already in the case, not targets (collection). |
Batch Execution (1 tool)
| Tool | Description |
|------|-------------|
| batch_scan | Run a tool against all files in a directory with safety bounds (max_files cap, timeout, per-tool limits like 100 for capa) |
Generic Execution (1 tool)
| Tool | Description |
|------|-------------|
| run_windows_command | Execute any cataloged tool with arguments (catalog-gated) |
All per-tool wrappers (Zimmerman suite, Hayabusa, mactime) are consolidated into run_windows_command. The tool catalog defines each binary's input flags, output format, and FK knowledge mapping. batch_scan extends this to directory-level operations with safety caps.
Tool Catalog
Tools are defined in YAML catalog files under data/catalog/. The catalog currently contains 25 tool entries across 6 files:
| File | Count | Tools |
|------|-------|-------|
| zimmerman.yaml | 14 | AmcacheParser, AppCompatCacheParser, EvtxECmd, JLECmd, LECmd, MFTECmd, PECmd, RBCmd, RECmd, SBECmd, SQLECmd, SrumECmd, WxTCmd, bstrings |
| sysinternals.yaml | 3 | autorunsc, sigcheck, strings |
| memory.yaml | 1 | winpmem |
| timeline.yaml | 3 | Hayabusa, chainsaw, mactime |
| analysis.yaml | 3 | capa, yara, densityscout |
| collection.yaml | 1 | KAPE |
Each entry defines the binary name, input style, output format, timeout, FK knowledge name, install methods, and search paths:
# data/catalog/zimmerman.yaml (excerpt)
category: zimmerman
tools:
- name: AmcacheParser
binary: AmcacheParser.exe
description: "Parse Amcache.hve for program execution evidence"
input_flag: "-f"
output_format: csv
timeout_seconds: 300
fk_tool_name: AmcacheParser
install_methods:
- method: direct
url: "https://ericzimmerman.github.io/#!index.md"
- method: dotnet
command: "dotnet tool install --global AmcacheParser"
install_paths:
- "C:\\Tools\\ZimmermanTools"
local_work_folder: true (WxTCmd only) runs the tool, in a case, in a new private folder in the server account's temp folder, removed afterwards, on copies of its input file and of the -wal, -shm and -journal beside it, made in an input folder there. The tool loads a DLL it writes where it runs, which the case share doesn't allow; SQLite can't open the share's path; and it rewrites the files it opens, so it never opens the case's. Its output still goes to the case. It must be given full paths, its input flag once, with a file, and no other path but an output flag's value. The input's record lists the companions it copied, and the reply's copy_note names the copies.
Response Envelope
Every tool response is wrapped in a structured envelope with forensic-knowledge enrichment:
{
"success": true,
"tool": "run_windows_command",
"data": {"output": {"rows": ["..."], "total_rows": 42}},
"data_provenance": "tool_output_may_contain_untrusted_evidence",
"output_format": "parsed_csv",
"audit_id": "wintools-steve-20260220-01khxqe4jr7wn9rmegfd2ed8e8",
"examiner": "steve",
"caveats": [
"Amcache entries indicate installation, not necessarily execution",
"Timestamps reflect installation time, not last run"
],
"advisories": ["Cross-reference with Prefetch for execution confirmation"],
"corroboration": {
"artifacts": ["prefetch", "shimcache"],
"tools": ["PECmd", "AppCompatCacheParser"]
},
"field_notes": {"KeyLastWriteTimestamp": "Last time the registry key was modified"},
"discipline_reminder": "Evidence is sovereign -- if results conflict with your hypothesis, revise the hypothesis, never reinterpret evidence to fit"
}
| Field | Source | Description |
|-------|--------|-------------|
| audit_id | Audit | Unique identifier (wintools-{examiner}-{YYYYMMDD}-{ULID}) |
| caveats | forensic-knowledge | Artifact-specific limitations and interpretation warnings |
| advisories | forensic-knowledge | Usage guidance and common misinterpretation corrections |
| corroboration | forensic-knowledge | Suggested cross-reference artifacts and tools |
| field_notes | forensic-knowledge | Timestamp field meanings from artifact definitions |
| discipline_reminder | Built-in | Rotating forensic methodology reminder (14 total, cycled per call) |
In a case, a run of a tool whose catalog entry has an output flag that writes no file has success: false, and error says so with what the tool printed.
Configuration
Environment Variables
| Variable | Default | Description |
|----------|---------|-------------|
| WINTOOLS_TIMEOUT | 600 | Default command timeout in seconds |
| WINTOOLS_HOST | 127.0.0.1 | HTTP server bind address |
| WINTOOLS_PORT | `46
Truncated for display — read the full file on GitHub.
Related Skills
taste-skill
116.8kAgent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages
Agent-Reach
112.3kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
ruflo
81.9k🌊 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
79.8kCompress 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.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
