What gets blocked is not baked into the server — it is whatever policy your `StrategyKey` points at, so two connections with different strategy keys behave differently on the same text. Everything runs on the vendor's side, which means the content you screen leaves your machine; that is the trade for not hosting a moderation model yourself.
A hosted MCP endpoint from ChuangsiAI that checks both what goes into a large language model and what comes out of it. Risk detection covers compliance, ethics and safety, and which rules apply is decided by a strategy identifier you send alongside your key.
- Content checked on the way in and on the way out, rather than only on the output side
- Risk identified across several dimensions — compliance, ethics and safety
- Rules you control: the `StrategyKey` header selects a policy you configure in the console
- A hosted SSE endpoint over HTTPS, so nothing runs on your machine
An account on the ChuangsiAI console. Create an API key under APIKey management and copy a strategy identifier from the policy centre, then connect to `https://mcp.chuangsiai.com/sse` with the key in the `Authorization` header and the strategy in `StrategyKey`. The transport is SSE and there is nothing to install.
One command plus a key — npx -y mcp-remote https://mcp.chuangsiai.com/sse, then supply credentials