The API key gets you in; the strategy key decides what counts as a risk. Two applications with the same credentials and different strategy keys behave completely differently, so treat the policy as configuration to version and review, not a one-time setup step. Screening both directions is the part worth having — user input is where prompt attacks arrive, and output is where they show up.
A content-safety layer for LLM applications. It inspects both what goes into the model and what comes out, and flags or blocks material that breaches the policy you have configured in the vendor's console.
- Checks on both directions — input and output are screened, not just the model's reply
- Risk detection across several dimensions, covering regulatory compliance, ethics and safety
- Low-latency screening over HTTPS, meant to sit inline rather than as a batch review afterwards
- A policy engine you configure yourself, so the rules and the protection strategy are yours rather than fixed
A hosted SSE endpoint at `https://mcp.chuangsiai.com/sse` — nothing to install, and the package is chuangsiai-mcp, version 0.1.0. Two headers: `Authorization` carrying an API key created in the console's API key section, and `StrategyKey` identifying the policy you built in the strategy centre. Both come from the vendor console.