# raccioly/docguard [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/raccioly/docguard  
**GitHub Stars:** 32  
**npm Downloads (last month):** 6729  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/raccioly-docguard

## Description
Deterministic documentation-drift detection for AI agents. Validates a project's canonical docs against its actual code — no LLM guessing; every finding has a stable code and a suggested fix. Five read-only tools over stdio: guard, score, explain, verify-claims, diagnose. npx -y docguard-cli mcp

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "docguard": {
    "command": "npx",
    "args": ["-y","docguard-cli"]
  }
}
```

## Documentation

## What raccioly/docguard MCP server does

The raccioly/docguard MCP server checks whether a project's canonical documentation matches evidence in its codebase. It is intended for documentation-led development and for repositories where AI agents rely on specifications, claims, references, and generated sections as working context.

Checks are deterministic rather than based on an LLM's interpretation. Findings have stable codes and include suggested fixes, giving an agent or developer a repeatable way to identify and address drift. The server also keeps unresolved or unverified claims visible instead of treating them as confirmed.

The project is listed in the Developer Tools category and is available as the `docguard-cli` package. The repository describes it as a community extension for GitHub Spec Kit, where it validates artifacts produced during the specification workflow.

## How it works

The MCP process communicates over stdio. It reads the project supplied to a tool call, examines documentation and local repository evidence, and returns structured results without writing to the project. The five exposed tools are:

- `guard` validates documentation and can serve as an enforcement check.
- `score` calculates a documentation maturity grade.
- `explain` provides details about findings or checks.
- `verify-claims` checks selected declarations against available local evidence.
- `diagnose` analyzes issues and returns prompts intended to help an AI repair documentation.

The underlying command-line project contains validators for supported facts, references, and generated sections. Exact declarations can be placed in `.docguard-evidence.json` for verification against current local evidence; statements without that evidence remain unverified.

## Setup and configuration

Run the server directly with Node.js using `npx -y docguard-cli mcp`. Node.js 18 or newer is required. A client configuration can start that command as a stdio MCP server. The README specifically mentions Claude and Cursor as MCP clients.

A container image is also published at `ghcr.io/raccioly/docguard:latest`. Mount the repository at `/workspace`, run the container with an interactive stdin, and pass `/workspace` as `projectDir` in tool calls or use the default working directory. The container's stdout is reserved for the JSON-RPC channel. Pin a version instead of using `latest` when running it in CI.

## Tools and capabilities

The raccioly/docguard MCP server is useful for checking documentation before an AI agent acts on repository context, reviewing canonical specifications against implementation evidence, and exposing stable findings to CI or repair workflows. Its read-only behavior makes it suitable for inspecting a mounted working tree without allowing the MCP process to alter files.

The broader CLI also supports commands such as `init`, `generate`, `diagnose`, `guard`, and `score`, but the MCP interface exposes only the five tools listed above. The package is distributed through npm, with a Python wrapper also available; the Python wrapper delegates to `npx` and still requires Node.js.

## Limitations and notes

Validation only covers supported facts and available repository evidence. A claim is not automatically verified merely because it appears in documentation. The server does not edit files, apply suggested fixes, or replace human judgment about requirements and architectural decisions. Its transport is stdio, so unrelated output must not be written to stdout when it is used as an MCP process.

_Full upstream README: https://allmcps.com/mcp/raccioly-docguard/readme_