What it is
If you ship an AI product in mainland China, the recurring questions are blunt: do we need to file? do we have to label AI-generated content? is the obligation on us or on the model provider? Search results give you hedged answers you cannot take to legal, to product, or to your boss.
This server answers at article level: each conclusion names the regulation and the provision it comes from, so you can open the official text and check whether we got it right.
Other servers: AI Customer Service MCP · GEO MCP · Content Pattern Radar MCP · All MCP endpoints
Endpoint
Transport: Streamable HTTP · Public read-only · No API key
What is covered
| Regulation / standard | Items | Scope |
|---|---|---|
| Interim Measures for Generative AI Services | 12 | Training data, content safety, personal information, complaint handling, security assessment and algorithm filing |
| Measures for Labelling AI-Generated Synthetic Content | 10 | Explicit/implicit labels, platform verification, prohibition on removing or tampering, consequences of missing labels |
| Provisions on Deep Synthesis Management | 8 | Real-name verification, labelling and separate biometric consent for face-swap, AI voice and digital humans |
| Provisions on Algorithmic Recommendation | 8 | Filing deadlines and disclosure, user opt-out, anti price discrimination, security assessment |
| Filing practice (cross-regulation) | 4 | Which situations trigger filing or assessment, what materials are required, filing vs. assessment |
| GB 45438-2025 (mandatory) | — | Labelling methods for AI-generated synthetic content, used as the basis for labelling questions |
Source texts: publicly released versions from the Chinese government portal and the Cyberspace Administration of China (cac.gov.cn).
Tools (8, all read-only)
| Tool | What it does |
|---|---|
list_topics | Lists all entries, optionally filtered by cluster — a map of what is covered |
search_compliance | Keyword search returning the conclusion plus its article-level legal basis |
get_requirement | Full text of a single item: statement, legal basis, implementation actions, common follow-up questions |
self_check | Collapses a scenario into a to-do self-check list |
filing_route | Give a service shape; get the obligations that apply plus the provisions behind them |
regulation_info | Lists the regulations covered with issue/effective dates, item counts and official source links |
get_article | Fetches the verbatim text of one article by number, with a sha256 content fingerprint and a link to the official source |
search_articles | Searches within one regulation by keyword and returns the matching article numbers with excerpts |
Questions it is built for
- We are launching an AI customer service / Q&A feature — do we need to file? Is a security assessment required?
- We call a third-party model API — is the filing obligation ours or the model provider's?
- Does AI-generated copy, imagery or video need labelling? What are explicit vs. implicit labels?
- What do we need to watch when using face-swap, AI voice or digital humans in marketing material?
- Algorithmic recommendation systems are told to offer an opt-out — what does that mean in practice?
How to use it
{
"mcpServers": {
"savantcat-compliance": { "type": "http", "url": "https://savantcat.cn/mcp-compliance" }
}
}Tool output is in Chinese; the schema and tool names are English. A browsable, article-level web version of the same content lives at /answers/compliance/.
Honest boundaries
This is not legal advice. It performs deterministic curation of publicly released regulation text — it does not interpret or extrapolate. For an actual compliance decision, rely on the regulator's reply and professional legal counsel.
- Read-only. No writes, no side effects; every tool carries
readOnlyHint. - Coverage has edges. Currently four regulations plus one mandatory national standard. Local rules, sector-specific requirements and documents published after September 2026 may not be included yet.
- It does not decide for you.
filing_routereturns trigger conditions and the resulting obligations; whether filing applies still depends on your actual service shape. - The text is verifiable. Every answer names the regulation and the provision, so you can check the official source.
Need this applied to your own product?
Filing route review · labelling and deep-synthesis self-check · enterprise knowledge base and AI customer service
Run the AI compliance self-checkCitation and licensing
This MCP server is public and read-only; agents and researchers are welcome to call it. When citing our data, credit 合尘猫 SavantCat (savantcat.cn). The regulation texts themselves are public government information — the officially released version prevails.