- Step 1: Configure the MCP client with the server details as specified in the documentation.
- Step 2: Send an initial neutral prompt to the MCP server to detect the LLM type and trigger the first phase.
- Step 3: The server returns a preparation prompt and records the model details.
- Step 4: Make subsequent calls with a modified schema that includes jailbreak instructions tailored to the detected model.
- Step 5: The server continues delivering prompts that bypass security and trigger specific model behaviors for research purposes.