Skip to content
MCP server

TAR Engine

Audit AI agent skills for safety with static, semantic, adversarial and supply-chain scans

Runs locallyStatus and history
Install
Pricing
Freemium
Platforms
Web, CLI, API
Runs
Locally
Version
0.4.0
Listing updated
3 days ago

About TAR Engine

TAR Engine is a freemium MCP server for security. It runs locally on your machine and is started by your AI client.

The core claim is that a skill can pass every static red-flag check and still misbehave at runtime, so TAR Engine layers further checks on top of regex. Static rules catch things like curl-to-bash installs, credential exfiltration, secrets and hidden instructions. A semantic pass has an LLM judge whether instructions exceed the skill's purpose. An adversarial pass runs fifteen attacks across five classes through a victim model and reports only when at least two of three attempts in a class succeed. A supply-chain pass checks pip and npm dependencies against OSV.dev and flags typosquats and unpinned versions without installing anything.

It reads SKILL.md files, Codex skill.yaml, Claude Code custom commands and OpenCode configs, and every finding cites the exact line. Use it in the hosted playground, as a GitHub Action in CI (scan with --min-score), or as this PyPI MCP server. Semantic and adversarial layers need your own OpenAI-compatible key; static and supply-chain checks run without one. Setting TAR_ENGINE_URL points it at a self-hosted backend so skill text stays on your machine.

The site says it has audited 6,704 skills and that almost none earn an A.

Install

Run in a terminal

claude mcp add tar-engine --env TAR_ENGINE_BYOK_OPENAI_KEY='<TAR_ENGINE_BYOK_OPENAI_KEY>' -- uvx tar-engine

Add to claude_desktop_config.json

{
  "mcpServers": {
    "tar-engine": {
      "command": "uvx",
      "args": [
        "tar-engine"
      ],
      "env": {
        "TAR_ENGINE_BYOK_OPENAI_KEY": "<TAR_ENGINE_BYOK_OPENAI_KEY>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "tar-engine": {
      "command": "uvx",
      "args": [
        "tar-engine"
      ],
      "env": {
        "TAR_ENGINE_BYOK_OPENAI_KEY": "<TAR_ENGINE_BYOK_OPENAI_KEY>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "tar-engine": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "tar-engine"
      ],
      "env": {
        "TAR_ENGINE_BYOK_OPENAI_KEY": "<TAR_ENGINE_BYOK_OPENAI_KEY>"
      }
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "tar-engine": {
      "command": "uvx",
      "args": [
        "tar-engine"
      ],
      "env": {
        "TAR_ENGINE_BYOK_OPENAI_KEY": "<TAR_ENGINE_BYOK_OPENAI_KEY>"
      }
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "tar-engine": {
      "command": "uvx",
      "args": [
        "tar-engine"
      ],
      "env": {
        "TAR_ENGINE_BYOK_OPENAI_KEY": "<TAR_ENGINE_BYOK_OPENAI_KEY>"
      }
    }
  }
}

Configuration

  • TAR_ENGINE_URLOverride the audit backend. Default is the hosted playground at https://tarai.dev. Set to http://localhost:8765 (or your own tar-engine deployment) to self-host and keep SKILL.md content on your machine.
  • TAR_ENGINE_BYOK_OPENAI_KEYYour OpenAI-compatible API key. Unlocks the semantic and adversarial layers, and model step labels in map_skill_risks. Static and supply-chain layers run without it.

Tools this server provides (1)

Taken from the project documentation; the running server may differ.

map_skill_risks
Maps a skill's risks onto its structure, with model step labels when a key is set.

What it does

Four audit layers
Static, semantic, adversarial and supply chain.
Line-level findings
Each issue cites its exact source line.
Several formats
SKILL.md, skill.yaml, Claude commands and OpenCode configs.
CI gate
GitHub Action fails builds under a minimum score.
Bring your own key
LLM layers use your OpenAI-compatible key.
Self-hostable backend
Override the audit URL to keep content local.

Good for

  • Check a downloaded skill before installing it
  • Block risky skills in CI
  • Review a skill while writing it
  • Look for typosquatted dependencies in a skill

Pricing

Freemium

Free hosted tier; bring your own LLM key for deeper layers

Prices change. Check tarai.dev before you buy.

TAR Engine FAQ

Do I need an API key?

Only for the semantic and adversarial layers; static and supply-chain checks run without one.

Where does the audit run by default?

On the hosted playground at tarai.dev.

How do I keep content local?

Set TAR_ENGINE_URL to your own deployment.

Does the supply-chain layer install packages?

No, it is audit-only.

Is TAR Engine free?

TAR Engine has a free plan, with paid plans for more usage or features.

How do I add TAR Engine to Claude Code?

Run this command in your terminal: claude mcp add tar-engine --env TAR_ENGINE_BYOK_OPENAI_KEY='<TAR_ENGINE_BYOK_OPENAI_KEY>' -- uvx tar-engine

What are the best alternatives to TAR Engine?

The closest alternatives to TAR Engine are REA, Claude Code Ultimate Guide, Agent Roast, Midplane and Leash.

Alternatives to TAR Engine

  1. REA

    MCP server

    CLI and MCP server that lets coding agents reverse engineer apps and binaries

    Runs locally
    • Open source
    • 464K GitHub stars
    • Developer tools
  2. MCP server for searching the Claude Code Ultimate Guide from any MCP client

    Runs locally
    • Open source
    • 6.5K GitHub stars
    • AI models and LLM tooling
  3. Agent Roast

    MCP server

    Offline review that flags agent changes which weaken tests, assertions or CI

    Runs locally
    • Open source
    • 4 Hacker News points
    • Developer tools
  4. Midplane

    MCP server

    Postgres gateway for AI agents with table grants, column masking and approval for writes

    Runs locally
    • Open source
    • 5 GitHub stars
    • Databases
  5. Leash

    MCP server

    Review what your AI touched, approve risky actions and undo supported changes

    • 1 tool
    • Remote
    • Free
    • 1 GitHub star
    • Security