five-mcp 1.0.1

160,000 deductively-derived JSON constraints that eliminate LLM persona drift. Measured: 0 breaks in a 120-turn pressure test.

five-mcp

MCP server for the FIVE Persona Engine — an LLM persona constraint engine that generates structured JSON constraints to eliminate persona drift. Instead of describing personality in words (which LLMs interpret differently each turn), FIVE defines behavioral parameters the LLM executes as a recipe. See how it works →

Measured: with the constraint JSON + free harness, the demo character survived a 120-turn pressure test with zero persona breaks (plain prompt: 8 breaks; JSON alone: 1). Numbers, transcripts and scripts →

Quick Start

Install

pip install five-mcp

Configure

Set your API key as an environment variable:

export FIVE_API_KEY=five_sk_your_key_here

Get your key at fiveengine.dev.

Use with Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "five-character-engine": {
      "command": "five-mcp",
      "env": {
        "FIVE_API_KEY": "five_sk_your_key_here"
      }
    }
  }
}

Use with other MCP clients

Any MCP-compatible client can connect via stdio transport:

five-mcp

Tool: generate

Generates persona constraints via the FIVE engine.

Parameters

Parameter	Type	Required	Description
character_name	string	Yes	Name of the character
q1 – q4	A / B / C / D	Yes	Personality axis choices
s1 – s4	1–5	No	Style sliders (default: 3)
free_text	string	No	Free-form description

Response

{
  "status": "ok",
  "remaining": 42,
  "constraint": { "..." }
}

Pricing

Each generate call costs $1 and consumes one credit. Manage credits at fiveengine.dev.

Links

• API & Docs: fiveengine.dev
• GitHub: github.com/kiro0x/five-mcp
• Design philosophy & examples: five-character-engine README
• Measured evaluation (120-turn drift test): five-character-engine eval/
• Engine repo: github.com/kiro0x/five-character-engine

License

MIT