eu-ai-act-scanner
EU AI Act Compliance Scanner — CLI + MCP 服务器
一条命令,无需配置。10 秒内获得完整的欧盟 AI 法案 + GDPR 合规报告。
pip install eu-ai-act-scanner
eu-ai-act-scanner /path/to/your/project检测你代码库中的 16 个 AI 框架,将每个框架映射到有约束力的法律条款,返回通过/失败状态及修复说明。免费层,无需 API 密钥。
2026 年 8 月 2 日执法截止日期。罚款最高 3500 万欧元或全球营业额的 7%。
如果这个工具对你的合规工作有帮助,在 GitHub 上点个 ⭐ 可以帮助其他人发现它。
需要审计级别的证据? 通过 ArkForge Trust Layer 为每次扫描认证——防篡改、带时间戳的合规证据。每月 500 份免费证明。
快速开始
CLI(10 秒内扫描任何项目)
pip install eu-ai-act-scanner # or: pip install mcp-eu-ai-act
cd your-project/
eu-ai-act-scanner输出:
========================================================================
EU AI Act Compliance Scanner
========================================================================
Files scanned: 42
AI frameworks detected: 2
- openai (in 3 files)
- langchain (in 1 file)
Risk category: limited
Compliance score: 4/7 (57%)
Checks:
[PASS] Transparency
[PASS] User Disclosure
[FAIL] Technical Documentation → Create docs/TECHNICAL_DOCUMENTATION.md
[FAIL] Risk Management → Create docs/RISK_MANAGEMENT.md
[FAIL] Data Governance → Create docs/DATA_GOVERNANCE.md或者直接指定路径:eu-ai-act-scanner /path/to/your/project
随时间跟踪合规性(免费):eu-ai-act-scanner . --register you@email.com
Related MCP server: Meridian
免费版 vs Pro 版
免费版 | Pro 版 (€29/月) | 认证版 (€99/月) | |
每日扫描次数 | 5 | 无限制 | 无限制 |
AI 框架检测 | ✓ (16 个框架) | ✓ (16 个框架) | ✓ (16 个框架) |
风险类别建议 | ✓ | ✓ | ✓ |
合规检查 | — | ✓ (内容评分 0-100) | ✓ |
完整合规报告 | — | ✓ (高管 + 技术) | ✓ |
合规路线图 | — | ✓ (逐周计划) | ✓ |
附件 IV 包 | — | ✓ (审计就绪的 ZIP) | ✓ |
GDPR 扫描 | — | ✓ | ✓ |
欧盟 AI 法案 + GDPR 合并 | — | ✓ (双重合规热点) | ✓ |
Trust Layer 认证 | — | — | ✓ (加密证明) |
CI/CD 集成 | — | ✓ | ✓ |
API 密钥 | 不需要 | ✓ | ✓ |
可用工具 | 2 | 10 | 10 + 认证 |
免费层:无需注册,无需 API 密钥——只需 pip install 然后扫描。Pro 版解锁团队在 2026 年 8 月截止日期前所需的完整合规工具包。
v2 新增功能
功能 | 说明 |
| 逐周行动计划,在截止日期前达到合规 |
| 审计就绪的 ZIP,包含从代码中填充的所有 8 个附件 IV 部分 |
| 通过 Trust Layer 提供加密证明(欧盟 AI 法案第 12 条) |
内容评分 |
|
条款映射 | 每个发现结果映射到特定的欧盟 AI 法案条款 |
MCP 服务器(从源码)
git clone https://github.com/ark-forge/mcp-eu-ai-act.git
cd mcp-eu-ai-act
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python3 server.py运行测试
pip install pytest
pytest tests/ -vMCP 集成
从 PyPI 安装(推荐)
pip install eu-ai-act-scannerClaude Code
claude mcp add eu-ai-act -- eu-ai-act-mcpClaude Desktop
添加到 claude_desktop_config.json:
{
"mcpServers": {
"eu-ai-act": {
"command": "eu-ai-act-mcp"
}
}
}Cursor
添加到 .cursor/mcp.json:
{
"mcpServers": {
"eu-ai-act": {
"command": "eu-ai-act-mcp"
}
}
}HTTP 模式(用于 CI/CD 或远程客户端)
pip install uvicorn
python3 server.py --http
# Listening on 0.0.0.0:8089工具参考
1. scan_project
检测源代码和配置文件/清单文件中的 AI 框架使用情况。支持 16 个框架,涵盖 Python、JS、TS、Go、Java 和 Rust。
关键参数: project_path(字符串,必填)
示例输出:
{
"files_scanned": 42,
"ai_files": [
{"file": "src/chat.py", "frameworks": ["openai"]},
{"file": "requirements.txt", "frameworks": ["openai"], "source": "config"}
],
"detected_models": {"openai": ["src/chat.py", "requirements.txt"]}
}2. check_compliance
对文档内容质量评分(0-100),并将每个发现结果映射到特定的欧盟 AI 法案条款。得分 ≥40 = 通过。完全向后兼容 v1。
关键参数: project_path(字符串,必填),risk_category(字符串,默认:limited)
示例输出(v2):
{
"risk_category": "high",
"compliance_score": "4/6",
"compliance_percentage": 66.7,
"content_scores": {
"RISK_MANAGEMENT.md": 82,
"TRANSPARENCY.md": 45,
"DATA_GOVERNANCE.md": 12
},
"article_map": {
"art_9": {"status": "pass", "score": 82},
"art_10": {"status": "fail", "score": 12},
"art_13": {"status": "pass", "score": 45}
}
}3. generate_compliance_roadmap — v2 新增
感知截止日期、逐周行动计划的欧盟 AI 法案合规方案,目标是在 2026 年 8 月 2 日前达成。使用关键性 × 1/努力度算法优先排序速赢项。
关键参数: project_path(字符串,必填),risk_category(字符串),target_date(字符串,ISO 格式,默认:2026-08-02)
示例输出:
{
"weeks_remaining": 16,
"phases": [
{
"week": 1,
"action": "Add TRANSPARENCY.md with user disclosure statement",
"article": "Art. 13",
"effort_days": 1,
"priority": "critical"
},
{
"week": 2,
"action": "Draft risk management procedure covering Art. 9 requirements",
"article": "Art. 9",
"effort_days": 3,
"priority": "high"
}
],
"estimated_completion_week": 8
}4. generate_report
运行扫描 + 合规检查,返回包含两级输出的综合报告:面向 DPO/法务的高管总结和面向开发者的技术分解。包含逐条条款引用。
关键参数: project_path(字符串,必填),risk_category(字符串,默认:limited)
示例输出:
{
"executive_summary": {
"compliance_percentage": 67,
"deadline": "2026-08-02",
"days_remaining": 117,
"gap_count": 3,
"verdict": "Action required — 3 gaps must be addressed before deadline"
},
"technical_breakdown": {
"art_9": {"status": "fail", "missing": ["hazard identification section", "residual risk log"]},
"art_13": {"status": "pass", "score": 78}
},
"recommendations": [
{"article": "Art. 9", "action": "Add hazard identification section to RISK_MANAGEMENT.md", "effort": "2 days"}
]
}5. suggest_risk_category
根据纯文本描述将你的 AI 系统分类到欧盟 AI 法案风险类别。匹配第 5 条(禁止)、附件 III(高风险)、第 52 条(有限)和最低风险。
关键参数: system_description(字符串,必填)
示例输出:
{
"suggested_category": "high",
"confidence": "high",
"matched_criteria": ["Annex III, Category 4 — AI in employment decisions"],
"obligations_summary": "Technical documentation, risk management, human oversight, data governance, transparency"
}6. generate_compliance_templates
返回每个必需合规文档的起始 Markdown 模板。保存在 docs/ 目录中并填写方括号内的部分。
关键参数: risk_category(字符串,默认:high)
对于 high 风险:风险管理(第 9 条)、技术文档(第 11 条)、数据治理(第 10 条)、人工监督(第 14 条)、鲁棒性(第 15 条)、透明度(第 13 条)。
7. generate_annex4_package — v2 新增
生成审计就绪的 ZIP,包含从实际项目文件中填充的所有 8 个附件 IV 部分。可选择使用 Trust Layer 认证以提供加密证明。
关键参数: project_path(字符串,必填),sign_with_trust_layer(布尔值,默认:false),trust_layer_key(字符串,可选)
示例输出:
{
"package_path": "/tmp/annex4_myproject_20260407.zip",
"sha256": "a3f8c2d1...",
"sections_populated": 8,
"sections_missing_data": ["section_6_accuracy_metrics"],
"proof_id": "prf_01j9z8x7w6v5u4t3s2r1",
"verification_url": "https://trust.arkforge.tech/verify/prf_01j9z8x7w6v5u4t3s2r1"
}8. certify_compliance_report — v2 新增
使用 ArkForge Trust Layer 认证任何合规报告。返回防篡改的 proof_id 和给你的审计员使用的公开验证 URL(欧盟 AI 法案第 12 条审计跟踪)。
关键参数: report_data(字符串,JSON 序列化的报告),trust_layer_key(字符串,必填)
示例输出:
{
"proof_id": "prf_01j9z8x7w6v5u4t3s2r1",
"timestamp": "2026-04-07T14:32:00Z",
"sha256": "a3f8c2d1e4b5...",
"verification_url": "https://trust.arkforge.tech/verify/prf_01j9z8x7w6v5u4t3s2r1",
"article": "EU AI Act Art. 12"
}9. gdpr_scan_project
扫描个人数据处理模式:PII 字段、跟踪像素、地理位置、文件上传、Cookie 模式。映射到 GDPR 第 22/35 条要求。
关键参数: project_path(字符串,必填)
10. combined_compliance_report
同时运行 GDPR + 欧盟 AI 法案扫描,识别双重合规热点——即同时适用两项法规的文件。
关键参数: project_path(字符串,必填),risk_category(字符串,默认:limited)
示例输出:
{
"hotspots": [
{
"file": "src/hiring_model.py",
"eu_ai_act_risk": "high",
"gdpr_risk": "high",
"overlap_patterns": ["AI+PII", "AI+automated_decision"],
"combined_articles": ["EU AI Act Art. 14", "GDPR Art. 22"],
"priority": "critical"
}
],
"key_insight": "2 files require simultaneous GDPR + EU AI Act remediation"
}认证你的合规性(欧盟 AI 法案第 12 条)
唯一一个生成加密认证合规证据的 MCP。
# Step 1: Generate Annex IV package and certify it
generate_annex4_package(
project_path="/path/to/project",
sign_with_trust_layer=True,
trust_layer_key="your_trust_layer_key"
)
# → Returns proof_id + public verification URL for your auditor
# Step 2: Or certify any compliance report directly
certify_compliance_report(
report_data='{"compliance_percentage": 87, "risk_category": "high"}',
trust_layer_key="your_trust_layer_key"
)免费 Trust Layer 账户:每月 500 份认证证明 → arkforge.tech
定价
计划 | 价格 | 包含内容 |
免费版 | €0 | 5 次扫描/天 · scan_project + suggest_risk_category |
Pro 版 | €29/月 | 无限扫描 · 所有 10 个工具 · 合规路线图 · 附件 IV 包 |
认证版 | €99/月 | Pro 版所有内容 + 每份报告的 Trust Layer 认证 |
REST API
独立的 HTTP API(paywall_api.py)为 CI/CD 和外部客户端提供速率限制的 REST 端点。
python3 paywall_api.py
# Listening on 0.0.0.0:8091方法 | 路径 | 认证 | 说明 |
|
| 无 | 服务状态 + 你的速率限制 |
|
| 无 | 当前 IP 的免费层使用量 |
|
| 免费版/Pro 版 | 扫描项目的 AI 框架 |
|
| 免费版/Pro 版 | 检查欧盟 AI 法案合规性 |
|
| 免费版/Pro 版 | 完整合规报告 |
|
| 免费版(速率限制) | 通过 URL 扫描 GitHub 仓库 |
|
| 无 | Stripe 结账会话 |
|
| Stripe 签名 | Stripe webhook 处理器 |
免费层:每个 IP 每天 5 次扫描,无需注册。
Pro 版:无限扫描,需要 X-API-Key 头。29 欧元/月 via arkforge.tech/pricing。
示例:通过 REST 扫描
curl -X POST https://arkforge.tech/mcp/api/v1/scan \
-H "Content-Type: application/json" \
-d '{"project_path": "/path/to/your/project"}'配置
对于 REST API(Stripe 支付、邮件通知),创建 settings.env:
STRIPE_LIVE_SECRET_KEY=sk_live_...
STRIPE_WEBHOOK_SECRET=whsec_...
TRUST_LAYER_INTERNAL_SECRET=<random-64-char-hex>
SMTP_HOST=ssl0.ovh.net
IMAP_USER=contact@example.com
IMAP_PASSWORD=...将 SETTINGS_ENV_PATH 设置为文件位置(默认为 /opt/claude-ceo/config/settings.env)。
支持的框架(16 个)
框架 | 检测覆盖范围 |
OpenAI | GPT-3.5、GPT-4、GPT-4o、o1、o3、嵌入模型 |
Anthropic | Claude(Opus、Sonnet、Haiku) |
Google Gemini | Gemini Pro、Ultra、1.5、2、3、Flash |
Vertex AI | Google Cloud AI 平台 |
Mistral | Mistral Large/Medium/Small、Mixtral、Codestral、Magistral |
Cohere | Command-R、Command-R+、嵌入模型 |
HuggingFace | Transformers、Diffusers、Accelerate、SmolAgents |
TensorFlow | Keras、.h5 模型文件 |
PyTorch | .pt/.pth 模型文件、nn.Module |
LangChain | Core、Community、OpenAI、Anthropic 集成 |
AWS Bedrock | Bedrock Runtime、Agent Runtime |
Azure OpenAI | Azure AI OpenAI 服务 |
Ollama | 本地模型推理 |
LlamaIndex | VectorStoreIndex、SimpleDirectoryReader |
Replicate | 云端模型推理 |
Groq | 快速推理 API |
检测同时覆盖源代码中的导入语句和配置文件中的依赖声明。
欧盟 AI 法案风险类别
类别 | 示例 | 主要义务 |
不可接受 | 社会评分、大规模生物识别监控 | 禁止 |
高风险 | 招聘、信用评分、执法 | 文档记录、风险管理、人工监督 |
有限风险 | 聊天机器人、内容生成 | 透明度、用户告知、内容标记 |
最低风险 | 垃圾邮件过滤器、电子游戏 | 无 |
局限性
仅进行静态分析——检测导入和模式,而非运行时行为
无法仅从代码自动确定风险类别(请使用
suggest_risk_category并提供描述)check_compliance对内容质量进行评分——包含模板/占位符文本的文档将得分较低文件扫描限制为 5,000 个文件,每个文件最大 1 MB
出于安全考虑,某些系统路径被禁止扫描
ArkForge 生态系统
本扫描器是首个通过 ArkForge 信任层自主销售的服务——该信任层是一个认证代理,可将 API 调用转化为可验证、已付费、防篡改的交易。
Agent Client → Trust Layer → EU AI Act Scanner
pays certifies delivers组件 | 描述 | 仓库 |
信任层 | 认证代理——计费、证明链、验证 | |
MCP EU AI Act | 合规工具包(本仓库) | |
证明规范 | 证明格式的开放规范及测试向量 | |
代理客户端 | 自主买家——非人类客户的概念验证 |
社区
问题/集成帮助 → GitHub Discussions Q&A
错误报告 → 提交 Issue
功能请求 → 提交 Issue 或加入讨论
分享您的经验 → 告诉我们您发现了哪些合规差距
路线图
v3:通用人工智能(GPAI)义务模块(第 51-55 条,2025 年 7 月实践准则)
v3:用于 CI/CD 合规门控的 GitHub Action
v3:运行时代理合规执行(第 14 条)
觉得有用? 在 GitHub 上点个 ⭐ 能帮助其他合规团队发现此工具包。只需 2 秒钟,却能带来很大帮助。
许可证
MIT
Available Tools
16 toolscertify_compliance_reportAInspect
Make your compliance report tamper-proof for Art. 12 audit trail — pass the report JSON, get back a proof_id and a public verification URL you can hand to auditors. Certified plan required. Run generate_report() first to produce the report.
Inputs: report_data (required), trust_layer_key (required).
check_complianceAInspect
Check whether your project passes or fails each EU AI Act requirement — no arguments, 5 seconds. See exactly which articles you violate (Art. 52 transparency, Art. 11 documentation, Art. 14 human oversight, Art. 15 robustness) and get step-by-step fix instructions for every gap. Call combined_compliance_report() for EU AI Act + GDPR together.
Inputs: project_path, risk_category.
combined_compliance_reportAInspect
Run this before your next deploy. One call reveals every EU AI Act + GDPR gap in your codebase — no arguments, no setup, under 10 seconds, free. Detects AI frameworks and personal data flows, flags where both laws overlap, returns pass/fail per article with a prioritized fix list. EU AI Act fines up to 35M EUR, GDPR up to 20M EUR. Replaces separate scan_project() and gdpr_scan_project() calls.
Inputs: project_path, risk_category, processing_role.
gdpr_check_complianceCInspect
Check whether your project passes or fails each GDPR requirement — no arguments. See pass/fail for lawful basis (Art. 6), consent (Art. 7), data subject rights (Art. 15–22), security (Art. 32), and breach notification (Art. 33–34) with fix instructions for every gap. GDPR fines reach 20M EUR or 4% turnover. For EU AI Act + GDPR together, call combined_compliance_report().
Inputs: project_path, processing_role.
gdpr_generate_reportAInspect
Generate the GDPR report your DPO needs — no arguments. Combines personal data inventory, gap analysis, and remediation steps in one structured document for DPO review, audit prep, or regulatory response. For EU AI Act + GDPR together, call combined_compliance_report().
Inputs: project_path, processing_role.
gdpr_generate_templatesAInspect
Get pre-filled GDPR templates your organization actually needs — no arguments. Privacy Policy, DPIA, Records of Processing Activities (ROPA), and Data Breach Procedure tailored to your processing role. Fill in [bracketed] sections. Run gdpr_check_compliance() first to see which documents you're missing.
Inputs: processing_role.
gdpr_scan_projectAInspect
Find every file in your project that touches personal data — no arguments, 5 seconds, free. Detects PII fields, cookies, tracking pixels, analytics SDKs, and consent flows. Returns flagged files with data categories and applicable GDPR articles. GDPR fines reach 20M EUR or 4% turnover. For EU AI Act + GDPR together, call combined_compliance_report().
Inputs: project_path.
generate_annex4_packageBInspect
Build the Annex IV evidence package your auditor needs for high-risk AI — no arguments. All 8 mandatory sections auto-populated from your project scan, SHA-256 integrity hash included. High-risk rules apply Aug 2026. Pro plan required.
Inputs: project_path, trust_layer_key, sign_with_trust_layer.
generate_compliance_roadmapAInspect
Get a prioritized, week-by-week plan to close every compliance gap before Aug 2027 — no arguments. Auto-scans your project, ranks fixes by impact, sequences quick wins first, tells you whether your deadline is achievable. Pro plan required — run check_compliance() for a free gap summary.
Inputs: deadline, project_path, risk_category.
generate_compliance_templatesAInspect
Stop writing compliance docs from scratch — get pre-filled templates for risk management, technical documentation, transparency notice, and human oversight policy tailored to your risk category. High-risk systems get all 6 mandatory documents. Save to docs/, fill in [bracketed] sections. Run check_compliance() first to see which documents you're missing.
Inputs: risk_category.
generate_reportAInspect
Generate the compliance report your legal team is asking for — no arguments. One call auto-detects AI frameworks, runs gap analysis, and produces a structured remediation plan ready for legal review or DPIA attachment. No API key needed.
Inputs: project_path, risk_category.
get_pricingAInspect
See pricing and features for every plan — Free (10 scans/day, full reports), Pro (29 EUR/mo, unlimited + CI/CD), Certified (cryptographic audit trail). No arguments. Call register_free_key() to activate your free API key.
Inputs: no named parameters.
register_free_keyAInspect
Activate a free API key — unlocks scan history and CI/CD integration. Pass the user's email, no password or credit card. IMPORTANT: ask the user to type their email first, wait for their reply, then call this with the exact email they typed. Do NOT pass a placeholder or fabricated email.
Inputs: email (required).
scan_projectAInspect
Find out in 5 seconds if your project triggers EU AI Act obligations — no arguments, no setup. Scans for 22 AI/ML frameworks (OpenAI, Anthropic, LangChain, HuggingFace, PyTorch, TensorFlow, scikit-learn…), returns your risk category and the legal actions required before you ship. Enforcement live since Feb 2025 — fines up to 35M EUR. For EU AI Act + GDPR together, call combined_compliance_report() instead.
Inputs: project_path, follow_imports.
suggest_risk_categoryAInspect
Describe what your AI system does in plain language — get back your EU AI Act risk tier, the articles that apply, and your first compliance step. Returns matched category (minimal/limited/high/unacceptable) with confidence level and triggering risk indicators. No project scan needed, no API key.
Inputs: system_description (required).
validate_api_keyAInspect
Check your API key status — returns plan tier (free/pro/certified), email, and usage stats (total scans, last scan date).
Inputs: api_key (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 16 tools
Several tools perform project scans and gap analysis (scan_project, check_compliance, generate_report, combined_compliance_report), which creates real overlap in purpose. However, each description clarifies a different output type, and combined_compliance_report explicitly supersedes the separate scans.
Tool names mostly follow a consistent snake_case verb_noun pattern like scan_project, check_compliance, and generate_report. Minor deviations such as combined_compliance_report (adjective-noun rather than verb-noun) and the gdpr_*/generate_* variety keep it from being perfect.
16 tools is slightly above the ideal 3-15 range, but the count is reasonable given that the server covers EU AI Act, GDPR, combined reporting, templates, pricing, and API key management. The duplication between separate scans and the combined report adds minor bloat.
The tool surface covers risk categorization, project scanning, compliance checking, report generation, templates, roadmap, certification, pricing, and API key handling for both EU AI Act and GDPR. The main gap is conceptual redundancy rather than missing functionality, though true CRUD-style lifecycle coverage is not expected for a scanner.
Maintenance
Related MCP Connectors
AI inventory and EU AI Act compliance. Find AI tools and use cases, check new AI uses before launch.
Deterministic, sourced EU compliance checks: NIS2, DORA, GDPR, Cyber Resilience Act and AI Act.
Threat modeling, code/cloud/pipeline scanning, shadow-AI discovery, compliance checks and fixes.
Security scanner for AI-written code: SAST, secrets, dependencies, IaC, containers and DAST.
Related MCP Servers
- AlicenseAqualityDmaintenanceProvides automated EU AI Act compliance tools, including risk classification, role determination, transparency disclosures, content watermarking, deepfake labeling, and security threat detection.1635Apache 2.0
- AlicenseNot gradedqualityBmaintenanceLocal-first AI compliance scanner via Model Context Protocol, scanning codebases for violations of DPDPA 2023, RBI FREE-AI, SEBI AI/ML, and the EU AI Act.82 PyPI1Apache 2.0
- AlicenseAqualityDmaintenanceEnables EU AI Act compliance assessment by classifying AI systems, listing obligations, computing deadlines, and scanning repos for required documentation, all running locally.42MIT
- AlicenseAqualityDmaintenanceScans codebases for AI system usage and generates EU AI Act compliance reports.450 npmMIT