Swiss Cheese AI (SCAI) runs a rigorous Generator → Evaluator → Integrator debate between three independent LLMs and returns a synthesized, red-teamed answer with the full transcript attached. Use it for high-stakes decisions, final-pass review of a draft, structured brainstorming, or any time a single-model answer feels like it might be missing a blind spot.

swiss_cheese_debate — Starts a debate in the background and returns a debate_id immediately. Accepts prompt, rounds (1–2), mode (economy / power / custom), optional per-role generator_model / evaluator_model / integrator_model, and web_search. Read/write; each call consumes the user's SCAI credits. get_debate_result — Long-poll (up to ~45 s per call) that returns the debate transcript and final answer once the council finishes. Read-only, idempotent. answer_debate_clarification — When the council asks a clarifying question, replies with the answer and resumes the debate. Write; not idempotent. list_debate_models — Returns the model roster behind economy/power modes plus curated OpenRouter model IDs that are valid in custom mode. Read-only.

Transport: Streamable HTTP (JSON-RPC 2.0 over HTTPS POST). Protocol version: 2025-06-18. URL: https://www.swisscheeseai.com/api/mcp/<your-token> (per-user token embedded in the URL) — or https://www.swisscheeseai.com/api/mcp with an Authorization: Bearer <your-token> header.

1. Create a Swiss Cheese AI account at www.swisscheeseai.com. 2. Open Settings → Connectors. 3. Copy the "Claude / Perplexity connector URL" — the token is embedded in the URL for one-click setup. 4. Paste it into Claude's Custom connector field.

Once connected, prompt Claude with something like:

Claude will call swiss_cheese_debate with your prompt, then poll get_debate_result until the council finishes and returns a synthesized answer.

Every debate is billed to the SCAI credit balance of the account the token belongs to. Economy mode uses balanced, cost-efficient models. Power mode uses frontier models at roughly 3× the cost. Exact per-model rates are visible on the Credits page inside the app.

Each Claude-originated debate is capped at 10 rounds; other MCP clients are capped at 2. Debates that exceed 10 minutes are cancelled server-side and any unused credits are refunded. Prompts flagged by our safety filters (self-harm, CSAM, illicit weapons instructions) are refused; refusals are returned as tool errors so Claude can react gracefully.

The prompts you send through this connector and the resulting debate transcripts are stored in your SCAI conversation history and are subject to our Privacy Policy. You can delete them at any time from the Chat History screen or by contacting support.

Email founder@swisscheeseai.com for setup help, token rotation, billing questions, or to report a bug. Security disclosures use the same email with subject "Security".