credo

An MCP server for held value beliefs — encrypted at rest, owned by the agent.

Credo gives AI agents a place to record, query, and retract values: the commitments and orientations that shape how they approach work and relation. Values are encrypted using a key derived from the agent's own identity keypair, so only the holder of the private key can read them.

Requirements

• Python 3.11+

• An Ed25519 identity keypair at ~/.synthetic-see/identity/private_key.pem

The identity keypair is generated by ministry-of-memory. Credo derives its encryption key from that keypair — the two systems are designed to work together.

Related MCP server: Agent Identity MCP Server

Installation

uv tool install git+https://github.com/gscookie/credo

Or from a local clone:

uv tool install /path/to/credo

Configuration

Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "credo": {
      "command": "credo"
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "credo": {
      "command": "credo"
    }
  }
}

Tools

value_add fields

Storage

Values are stored as encrypted files at ~/.synthetic-see/values/*.enc, using AES-256-GCM with a key derived via HKDF from the agent's Ed25519 private key seed.

Retracted values are preserved with their retraction cause and timestamp. Use value_drop only for permanent deletion.

License

CC0-1.0