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-mcp2. Add to Claude Desktop
Open your Claude Desktop config file:
Mac:
~/Library/Application Support/Claude/claude_desktop_config.jsonWindows:
%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 a Python code string for all 6 EU AI Act articles |
| Scan a single |
| Scan an entire project directory |
Analysis
Tool | What it does |
| Deep analysis using local fine-tuned Llama model (via Ollama) |
| Scan text for 15 prompt injection patterns |
| Classify a tool/function by risk level (CRITICAL/HIGH/MEDIUM/LOW) |
Remediation
Tool | What it does |
| Generate working code to add AIR trust layer to your agent |
| Get the specific fix for a failing article |
Documentation
Tool | What it does |
| Technical explanation of what each article requires |
| 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-v2The 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 pipelinesDevelopment
git clone https://github.com/airblackbox/air-blackbox-mcp
cd air-blackbox-mcp
pip install -e .
python3 -m air_blackbox_mcpLinks
Website: airblackbox.ai
Docs: docs.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.