Mind Protocol MCP server and runtime for AI agents. Graph physics, traversal, structured dialogues.

pip install mind-mcp

Or from source:

git clone https://github.com/mind-protocol/mind-mcp.git
cd mind-mcp
pip install -e .

# Initialize a project
mind init

# Start FalkorDB (default backend)
docker run -p 6379:6379 falkordb/falkordb

Creates .mind/ with:

• Protocol docs (PRINCIPLES.md, FRAMEWORK.md)
• Agent definitions in .mind/actors/{name}/CLAUDE.md
• Skills, procedures, state tracking
• Python runtime for physics, graph, traversal

Agents are discovered dynamically from .mind/actors/{name}/CLAUDE.md.

Selection uses graph physics: score = similarity * weight * energy

All IDs follow {TYPE}_{Name} pattern:

• AGENT_Witness, AGENT_Fixer
• TASK_FixAuth_a7b3
• MOMENT_Assignment_witness_c4d2

Add to .mcp.json:

{
  "mcpServers": {
    "mind": {
      "command": "python3",
      "args": ["-m", "mcp.server"],
      "cwd": "/path/to/mind-mcp"
    }
  }
}

mind init [--database falkordb|neo4j]  # Initialize .mind/
mind status                             # Show status
mind doctor                             # Health checks
mind work [path]                        # AI-assisted work

docker run -p 6379:6379 falkordb/falkordb

mind init --database neo4j

Configure in .env:

DATABASE_BACKEND=neo4j
NEO4J_URI=neo4j+s://xxx.databases.neo4j.io
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_password

.mind/
├── PRINCIPLES.md          # How to work
├── FRAMEWORK.md           # Navigation guide
├── actors/                # Agent prompts
│   ├── witness/CLAUDE.md
│   ├── fixer/CLAUDE.md
│   └── ...
├── skills/                # Executable capabilities
├── procedures/            # Structured dialogues
└── state/                 # SYNC files

• Python 3.10+
• FalkorDB or Neo4j
• Optional: OpenAI API key (embeddings)

MIT