Murder Mystery

You run a complete murder mystery game engine exposed as an MCP server. It generates procedural mysteries, lets you interrogate AI suspects, search for clues, and solve crimes through a rich set of MCP tools. This server architecture supports secure truth handling via an Oracle pattern, built-in suspect emotions, and a memory-enabled RAG layer to keep investigations coherent across conversations.

How to use

You interact with the Murder Mystery MCP Server using an MCP client. Start a new game, ask suspects about their alibis, search locations for clues, and gather evidence to make a formal accusation. The server provides a clean separation between what you can read (the game state, clues, and timelines) and what you can do (start games, interrogate, search, and accuse). Use the tools to drive the investigation and the memory system to cross-check statements and uncover contradictions.

How to install

Prerequisites: you need Python installed on your system and a working MCP client to connect to the server.

1. Create a dedicated directory for the server and navigate into it.

2. Run the server as described in the runtime example to start accepting MCP requests.

Additional setup and usage notes

Environment variables you may need to configure for full functionality include your API keys for language and image capabilities. The server uses these keys to generate content, images, and voices if you enable those features.

For local development and testing, you can start a plain Python-based MCP server process and connect your MCP client to it. The server exposes a set of tools for game flow, investigation, and image generation, described in the Tools section.

MCP configuration snippet

{
  "mcpServers": {
    "murder_mystery": {
      "type": "stdio",
      "command": "python",
      "args": ["-m", "murder_mystery_mcp"],
      "cwd": "/path/to/murder-mystery-mcp",
      "env": {
        "OPENAI_API_KEY": "your-key",
        "HF_TOKEN": "your-hf-token",
        "ELEVENLABS_API_KEY": "your-elevenlabs-key"
      }
    }
  }
}

Available tools

start_game

Start a new procedurally-generated murder mystery. Initializes a fresh victim, suspects, clues, and encounter graph.

get_game_state

Return the current game state including suspects, clues, progress, and timeline.

interrogate_suspect

Question a suspect with memory and emotion-aware responses. Suspects remember past conversations and adapt their answers.

search_location

Search a location for clues and evidence, returning newly discovered items and relevant context.

make_accusation

Formally accuse a suspect of the murder, triggering resolution logic and potential end-of-game conditions.

search_memory

Search past conversations for relevant statements, supporting evidence gathering.

find_contradictions

Detect contradictions between a suspect’s statements and their past testimonies.

get_cross_references

Find what other suspects said about a person to compare narratives.

get_investigation_hint

Provide a hint when you’re stuck in the investigation.

get_timeline

Show the investigation timeline with key events and statements.

generate_portrait

Create a portrait image of a suspect for visual context.

generate_scene

Generate an image of a location or scene to aid visualization.

generate_title_card

Create an opening scene title card for the mystery.