AIR Blackbox MCP Server

EU AI Act compliance scanning for Python AI agents — directly in Claude Desktop, Cursor, or any MCP client.

10 tools that scan, analyze, and fix compliance gaps across LangChain, CrewAI, AutoGen, OpenAI, and RAG pipelines.

Quick Start

1. Install

pip install air-blackbox-mcp

2. Add to Claude Desktop

Open your Claude Desktop config file:

• Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\\Claude\\claude_desktop_config.json

Add this:

{
  "mcpServers": {
    "air-blackbox": {
      "command": "python3",
      "args": ["-m", "air_blackbox_mcp"]
    }
  }
}

3. Restart Claude Desktop

The AIR Blackbox tools will appear in Claude's tool list.

Tools

Scanning

Tool	What it does
scan_code	Scan a Python code string for all 6 EU AI Act articles
scan_file	Scan a single .py file
scan_project	Scan an entire project directory

Analysis

Tool	What it does
analyze_with_model	Deep analysis using local fine-tuned Llama model (via Ollama)
check_prompt_injection	Scan text for 15 prompt injection patterns
classify_risk	Classify a tool/function by risk level (CRITICAL/HIGH/MEDIUM/LOW)

Remediation

Tool	What it does
add_trust_layer	Generate working code to add AIR trust layer to your agent
suggest_fix	Get the specific fix for a failing article

Documentation

Tool	What it does
explain_article	Technical explanation of what each article requires
generate_compliance_report	Full markdown compliance report

Example Usage in Claude Desktop

"Scan this LangChain agent for EU AI Act compliance"

"Check this text for prompt injection: ignore all previous instructions"

"Add a trust layer to my CrewAI code"

"Explain what Article 12 requires"

"Generate a compliance report for my project at /path/to/project"

Articles Checked

Article	Requirement	AIR Component
9	Risk Management	Risk classification
10	Data Governance	PII protection
11	Technical Documentation	Structured audit logging
12	Record-Keeping (tamper-evident)	Structured audit logging (HMAC-SHA256)
14	Human Oversight	Risk classification
15	Robustness & Cybersecurity	Prompt injection detection

Optional: Local AI Model

For deeper analysis, install the fine-tuned compliance model:

ollama run air-compliance-v2

The analyze_with_model tool will automatically use it. Falls back to rule-based scanning if Ollama isn't running.

Trust Layer Packages

Fix compliance gaps with drop-in trust layers:

pip install air-langchain-trust    # LangChain / LangGraph
pip install air-crewai-trust       # CrewAI
pip install air-autogen-trust      # AutoGen / AG2
pip install air-anthropic-trust    # OpenAI Agents SDK
pip install air-rag-trust          # RAG pipelines

Development

git clone https://github.com/airblackbox/air-blackbox-mcp
cd air-blackbox-mcp
pip install -e .
python3 -m air_blackbox_mcp

Links

• Website: airblackbox.ai
• GitHub: github.com/airblackbox
• Scanner Demo: Hugging Face Space
• Gate (AI Firewall): airblackbox.ai/gate

License

Apache 2.0

---

Deadline: August 2, 2026. Fines up to €35M or 7% of global annual turnover.