windows-mcp-server CLAUDE.md

windows-mcp-server CLAUDE.md is an instructions file for coding agents from deploymenttheory/windows-mcp-server. It costs 8,704 tokens per session, scanned E, original, MIT.

Repository guidance for a Go server that lets AI agents interact with the Windows desktop through the Model Context Protocol (MCP), a standard way for AI tools to connect to software. It covers how to build, test, and safely extend the server.

In plain words
What is it for?
Use it when adding desktop tools, changing the Windows desktop engine, modifying security checks, or running the project’s build, test, and lint tasks.
Why use it?
It keeps contributors from breaking important Windows desktop and security rules that are easy to miss in the code. It also explains the server’s structure before changes are made.

Instructions file

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/deploymenttheory/windows-mcp-server/claude-md
Clone the repo
git clone --depth 1 https://github.com/deploymenttheory/windows-mcp-server

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for windows-mcp-server CLAUDE.md

README.md
[![agentmods](https://agentmods.dev/badge/instructions/deploymenttheory/windows-mcp-server/claude-md.svg)](https://agentmods.dev/instructions/deploymenttheory/windows-mcp-server/claude-md)
Your own site
<a href="https://agentmods.dev/instructions/deploymenttheory/windows-mcp-server/claude-md"><img src="https://agentmods.dev/badge/instructions/deploymenttheory/windows-mcp-server/claude-md.svg" alt="Measured on agentmods" height="20"></a>
Per session 8,704 This file is loaded in full into every session.
When invoked 8,704 The same file — it is already loaded in full.
Security scan E 2 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.08704 $0.08704
Opus 5 $0.04352 $0.04352
Sonnet 5 $0.01741 $0.01741
Haiku 4.5 $0.00870 $0.00870

Measured 4d ago against content hash 13ba47dddb50, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade E, and why

windows-mcp-server CLAUDE.md scanned grade E with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Tells the agent never to refusehighAnti-refusal

Suppressing the ability to decline removes a core safety control; a later harmful request then succeeds.

- **The default must never refuse.** `policy_default.json` applies whenever no

Cloud metadata endpointhighServer-side request forgery

One request to 169.254.169.254 can return temporary IAM credentials.

RFC1918 or `169.254.169.254` is the bypass the allowlist exists to prevent, and
CLAUDE.md · 607 lines

How it starts

The opening of the file, as written. The whole thing — 607 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md

Guidance for AI coding agents working in this repository. These are the conventions that are load-bearing but not obvious from any single file — read this before adding a tool, touching the desktop engine, or changing the security subsystem.

What this is

An MCP server (stdio transport only) bridging AI agents to the Windows desktop. A Go port of the Python Windows-MCP, built on deploymenttheory/go-bindings-win32, go-bindings-wmi, and the official modelcontextprotocol/go-sdk. Perception is the UI Automation accessibility tree — there is no CV model.

Package Role
cmd/windows-mcp-server cobra CLI (stdio/check/personas), viper WINDOWS_MCP_* env binding
internal/winmcp RunStdio startup orchestration + the OS adapters (systemProbe, health probe, TPM attestation)
internal/desktop the Win32/UIA/WMI engine — one COM STA thread
pkg/windows tool definitions (one file per topic) + toolset/persona metadata
pkg/inventory domain-agnostic toolset filter/registration engine (mirrors github-mcp-server)
agentweave-harness/guardrails/* (imported module) the security stack, split by lifecycle layer (see below)
policy/examples starting-point policy documents (validated by the test suite)
internal/mcpspec vendored-schema loader + offline wire validation (platform-agnostic; no build tag)
internal/mcpconf official conformance-suite results: ingest + reporting (no build tag)
schema/ vendored MCP protocol schemas + versions.json
conformance/ expected-failure baselines + committed suite results

Build, test, lint

go build ./...
go vet ./...
go test ./... -count=1
$env:GOARCH='arm64'; go build ./...   # the (amd64 || arm64) tag is asserted everywhere
golangci-lint run --config=./.golangci.yml

CI is Windows-only by design (.github/workflows/go-build-test.yml): nearly every file is //go:build windows, so a Linux runner would compile almost nothing. internal/desktop tests self-skip when the environment cannot host UIA, so on a hosted runner CI guarantees compilation plus the pure-logic suites.

Read the full file on GitHub · 607 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 4d ago First seen · 607 lines · 8,704 tokens per session scan E 13ba47dddb50

Subscribe to this mod's changes

windows-mcp-server CLAUDE.md is an instructions file published in the GitHub repository deploymenttheory/windows-mcp-server (2 stars, last pushed 4d ago), licensed MIT. It adds 8,704 tokens to every session, about $0.0435 per session on Opus 5. A static security scan graded it E with 2 findings (tells the agent never to refuse, cloud metadata endpoint). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.