# PrufGate

> PrufGate is an MCP server for security. Read-only checks for fake code, hollow tests and unguarded routes in AI-written changes. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Platforms:** CLI, self-hosted
- **Repository:** https://github.com/guardianvigil-labs/prufgate
- **MCP Registry name:** io.github.GuardianVigil-Labs/prufgate
- **Categories:** [security](https://ai-powered.com/categories/security), [developer tools](https://ai-powered.com/categories/developer-tools)
- **Listing updated:** 2026-11-15

## About

PrufGate targets code that compiles and passes a quick review but does not do the job: a catch block returning a hard-coded success, a click handler that does nothing, a test that asserts true or mocks the function under test, a skip or ts-ignore slipped into a diff, an undeclared import, or a route that loads a record by ID without an ownership check.

It is deterministic Python with no network calls and no LLM unless you choose one, and it runs in seconds on a pull request. Six engines cover anti-fabrication, test integrity, a diff guard, a contract guard (undeclared imports, env vars missing from .env.example, calls to nonexistent API routes), BOLA and IDOR checks across many web frameworks, and a diff reviewer for issues like string-built SQL or disabled TLS verification. Output is SARIF for code scanning and Markdown for people.

The maintainers say it uses pattern and structure heuristics rather than a full parser, so it can miss things and occasionally flag correct code; findings name their rule and can be suppressed with an inline comment, exemptions or a baseline. The MCP server comes from the PyPI package (the prufgate-mcp command) and needs Python 3.9 or newer and Git.

## Features

- **Anti-fabrication checks:** Finds canned success on error paths, simulated returns, mock data in production and stub UI handlers.
- **Test integrity engine:** Flags tests with no assertions, unfailable assertions, or deleted and weakened assertions in a diff.
- **Diff guard:** Catches added skips, bare suppressions, fake exits and weakened lint or coverage settings.
- **Contract guard:** Reports undeclared imports, undocumented env vars and calls to API routes that do not exist.
- **BOLA and IDOR detection:** Finds handlers that fetch by client-supplied ID without auth or owner scoping.
- **SARIF and Markdown output:** Feeds code scanning and gives readable reports; exit code 2 means the run failed, never a pass.

## Good for

- Review a pull request written by a coding agent for fake or stubbed work
- Catch tests that were weakened to make a build pass
- Check a branch for undeclared dependencies before merge
- Spot routes that expose records without ownership checks
- Add a gate next to linters in CI or pre-commit

## Install

### Claude Code (Terminal)

```sh
claude mcp add prufgate -- uvx prufgate
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "prufgate": {
      "command": "uvx",
      "args": [
        "prufgate"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "prufgate": {
      "command": "uvx",
      "args": [
        "prufgate"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "prufgate": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "prufgate"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "prufgate": {
      "command": "uvx",
      "args": [
        "prufgate"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "prufgate": {
      "command": "uvx",
      "args": [
        "prufgate"
      ]
    }
  }
}
```

## FAQ

### What does it need to run?

Python 3.9 or newer and Git; install with pipx or pip, or run the prufgate.pyz zipapp.

### Does it call the network or an LLM?

No. It is deterministic Python with no network calls and no LLM unless you ask for one.

### Will it replace my linter or SAST tool?

No, it is meant to sit next to them.

### Can it be wrong?

Yes. It uses heuristics, will miss some things and may flag correct code, so findings can be suppressed with a recorded reason.

### What do the exit codes mean?

0 is clean, 1 is blocking findings, 2 means it could not run.

### How do I add PrufGate to Claude Code?

Run this command in your terminal: claude mcp add prufgate -- uvx prufgate

### What are the best alternatives to PrufGate?

The closest alternatives to PrufGate are Brevio Core, AI Guard, HefestoAI, Toolproof and Gadriel AI Security Harness.

## Where this comes from

- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [Brevio Core](https://ai-powered.com/mcp/brevio-core): Brevio Core is a free MCP server for developer tools.
- [AI Guard](https://ai-powered.com/mcp/ai-guard): AI Guard is an open-source MCP server for developer tools.
- [HefestoAI](https://ai-powered.com/mcp/hefestoai): HefestoAI is a freemium MCP server for developer tools.
- [Toolproof](https://ai-powered.com/mcp/toolproof): Toolproof is a free MCP server for security.
- [Gadriel AI Security Harness](https://ai-powered.com/mcp/gadriel-ai-security-harness): Gadriel AI Security Harness is a freemium MCP server for security.
- [SLF Guardian](https://ai-powered.com/mcp/slf-guardian): SLF Guardian is a freemium MCP server for security.
- [valvur](https://ai-powered.com/mcp/valvur): valvur is an open-source MCP server for security.
- [Penthera](https://ai-powered.com/mcp/penthera): Penthera is an open-source MCP server for security.

---
Source: https://ai-powered.com/mcp/prufgate
