Air Blackbox MCP
EU AI Act compliance scanner for Python AI agents.
Connector by airblackbox · GitHub account verified by the MCP Registry
Install
Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.
npx agentmarket install connector/airblackbox-air-blackbox-mcp
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this server can do and what it asks for, from its tool list, install config and code. Checked Oct 25, 2026.
- Tools
- 12 read · 1 write · 0 delete · 1 not yet classified
- Secrets
- None asked for
- Install scripts
- None found
- Runs as
- Python package
Tools by what they do 14
Read from the source code; the running server may list different tools.
Read 12 · Look things up without changing anything.
analyze_with_modelcheck_injectionclassify_riskcompliance_historyexplain_articlescan_biasscan_codescan_filescan_gdprscan_projectsuggest_fixvalidate_action
Write 1 · Create, change, send or run things.
add_trust_layer
Not yet classified 1 · We could not tell from the name and description.
generate_compliance_report
Network hosts it talks to 1
github.com· found in code
Files it touches
- Any path the agent passes to its tools
Tools (14)
Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].
add_trust_layerGenerate modified code WITH the AIR trust layer added. Given existing AI agent code and framework name, returns working, copy-paste-ready code with imports and setup for the AIR trust layer. Supports: langchain, crewai, autogen, openai, haystack (rag), adk, claude. If framework is empty, auto-detects from the code.
analyze_with_modelAnalyze code using the local air-compliance-v2 fine-tuned model via Ollama. If Ollama is running with the air-compliance-v2 model, sends the code for deep analysis that catches issues regex patterns miss. Falls back gracefully to rule-based scanning if Ollama isn't available.
check_injectionScan text for prompt injection attacks. Checks 15 weighted patterns including role override, jailbreak attempts, system prompt manipulation, safety bypass, privilege escalation, and more. Returns detected patterns, confidence score (0-1), and whether the text would be blocked by a guardrail.
classify_riskClassify a tool or function by EU AI Act risk level. Maps tool names to risk categories: - CRITICAL: shell, exec, delete, spawn (requires human approval) - HIGH: database, email, deploy, git_push (requires approval or audit) - MEDIUM: http_request, api_call (should be logged) - LOW: file_read, search, query (standard logging) Uses the same risk map as AIR ConsentGate.
compliance_historyView past scan results, trends, and compliance history. Available only with full air-blackbox SDK (pip install air-blackbox-mcp[full]). Actions: 'list': Show recent scans 'trend': Analyze compliance trend over time 'score': Get latest compliance score 'export': Export full audit trail Helps track compliance progress and identify patterns.
explain_articleGet a technical explanation of a specific EU AI Act article. Explains what Article 9, 10, 11, 12, 14, or 15 requires, mapped to specific AIR trust layer components that address each requirement.
generate_compliance_reportGenerate a full markdown compliance report for AI agent code. Scans the code and produces a formatted report with: - Framework detection results - Trust layer status - Per-article findings with severity - Fix recommendations - Overall compliance score
scan_biasScan code for AI bias and fairness issues. Available only with full air-blackbox SDK (pip install air-blackbox-mcp[full]). Checks for: - Disparate impact risks in decision logic - Training data bias indicators - Protected attribute handling - Fairness metric awareness - Bias mitigation strategies
scan_codeScan AI agent code for EU AI Act compliance (Articles 9-15). Accepts Python, Jupyter notebook JSON, TypeScript, JavaScript, Java or Kotlin when the air-blackbox SDK is installed (pip install air-blackbox-mcp[full]); the language is detected from the code, or pass `language` to state it. Without the SDK, the built-in scanner reads Python only. Detects frameworks (LangChain, CrewAI, AutoGen, OpenAI, Haystack, LlamaIndex, and their npm/JVM counterparts), checks for trust layer presence, and evaluates compliance across 6 articles: - Art 9: Risk Management (error handling, fallbacks) - Art 10: Da…
scan_fileScan a single source file for EU AI Act compliance. Python, .ipynb, TypeScript, JavaScript, Java or Kotlin with the SDK installed; Python only without it. Reads the file, detects framework and trust layer, then runs all compliance checks. Returns line count, findings, and score.
scan_gdprScan code for GDPR compliance issues (data processing, consent, etc.). Available only with full air-blackbox SDK (pip install air-blackbox-mcp[full]). Checks for: - Personal data handling without consent - Data retention policies - Right to erasure implementation - Data processing agreements - Cross-border data transfer safeguards
scan_projectRecursively scan all source files in a project directory. Walks the directory tree (skipping node_modules, .git, build output, etc.) and runs a combined compliance analysis. With the SDK installed this reads Python, notebooks, TypeScript, JavaScript, Java and Kotlin and gives the verdict `air-blackbox comply --scan` gives; without it, .py files only. Returns per-file info, aggregate findings, and project-level score.
suggest_fixGet specific code fix suggestions for a given EU AI Act article. Provide an article number (9, 10, 11, 12, 14, or 15) and optionally the framework name for framework-specific recommendations.
validate_actionValidate an agent action before execution (Article 14 compliance). Available only with full air-blackbox SDK (pip install air-blackbox-mcp[full]). Use this to check if an action should be executed, requires approval, or should be blocked. Implements ConsentGate logic. Args: action_type: 'tool_call', 'file_write', 'api_call', 'shell_exec', etc. tool_name: name of the tool/function being called risk_level: optional override ('CRITICAL', 'HIGH', 'MEDIUM', 'LOW') Returns: approval status, required policies, and remediation steps
Signals
- Stars
- 3
- Forks
- 2
- Open issues
- 1
- Weekly downloads
- 94
- Last commit
- Oct 15, 2026 (13 days ago)
- Last release
- Oct 15, 2026 (13 days ago)
- Last package publish
- Oct 15, 2026 (13 days ago)
From GitHub (fetched Oct 24, 2026) and the package registry.
Details
- Registry name
io.github.airblackbox/air-blackbox-mcp- Version
- 0.2.0
- Auth
- No auth
- Packages
- pypi air-blackbox-mcp
- License
- Apache-2.0
- Repository
- github.com/airblackbox/air-blackbox-mcp
- First seen
- Oct 24, 2026
- Last checked
- Oct 25, 2026
Same server elsewhere
Copies of this server published by others; search shows only one of them.
Cite this page
Air Blackbox MCP. Agent Mart. https://agentmart.dev/connectors/airblackbox-air-blackbox-mcp (last updated 2026-10-25)