SkillAgentSearch skills...

wintools-mcp

Valhuntir Windows forensic tool execution via MCP

Install / Use

claude mcp add AppliedIR -- npx -y github:AppliedIR/wintools-mcp

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

78/100

Supported Platforms

Claude Code
Claude Desktop

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.

Substance
30/30
Structure
20/20
Description
8/15
Adoption
4/20
Freshness
15/15

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 found

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.

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.

SkillScoreStarsUpdatedFormat
wintools-mcp (this skill)by AppliedIR7810todayMCP Server
taste-skillby Leonxlnx100116.8k8d agoCLAUDE.md
Agent-Reachby Panniantong100112.3k9d agoCLAUDE.md
rufloby ruvnet10081.9ktodayCLAUDE.md
headroomby headroomlabs-ai10079.8ktodayCLAUDE.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.

Valhuntir

Windows Tools MCP

CI License: MIT

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/-l and update-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, --rlc and 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

View on GitHub
GitHub Stars10
CategoryDevelopment
Updated21h ago
Forks1

Languages

Python

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low1 info