otto-mcp 1.0.0

MCP server for Otto cognitive safety layer - ThinkingMachines [He2025] compliant

Otto MCP Server

Model Context Protocol (MCP) server for Otto cognitive safety layer.

Overview

This MCP server exposes Otto's cognitive state management to any MCP-compatible client (Claude Desktop, Cursor, VS Code, etc.). It enables cross-tool safety gating and cognitive state awareness.

Installation

pip install otto-mcp

Or install from source:

cd Otto/packages/otto-mcp
pip install -e .

Configuration

Claude Desktop

Add to your Claude Desktop config (~/.config/claude-desktop/config.json on Linux/macOS or %APPDATA%\Claude\config.json on Windows):

{
  "mcpServers": {
    "otto": {
      "command": "otto-mcp"
    }
  }
}

Cursor / VS Code

Add to your MCP settings:

{
  "mcp": {
    "servers": {
      "otto": {
        "command": "otto-mcp"
      }
    }
  }
}

Tools

otto_status

Get current cognitive state.

Burnout: GREEN | Energy: medium | Max Depth: deep

{
  "burnout_level": "green",
  "energy_level": "medium",
  "momentum_phase": "building",
  "mode": "focused",
  "max_thinking_depth": "deep",
  "should_intervene": false,
  ...
}

otto_check

Check if an operation is safe given current state.

Input:

{
  "operation": "deep architecture analysis",
  "thinking_depth": "ultradeep"
}

Output:

ADJUST DEPTH: Burnout at ORANGE - depth capped at standard

{
  "operation": "deep architecture analysis",
  "requested_depth": "ultradeep",
  "allowed": false,
  "recommended_depth": "standard",
  "reason": "Burnout at ORANGE - depth capped at standard"
}

otto_calibrate

Set focus and urgency calibration.

Input:

{
  "focus_level": "locked_in",
  "urgency": "deadline"
}

otto_expert

Get recommended intervention expert for a message.

Input:

{
  "message": "I'm so frustrated, nothing is working!"
}

Output:

Expert: VALIDATOR (priority 1)
Trigger: frustrated_detected

{
  "expert": "validator",
  "trigger": "frustrated_detected",
  "priority": 1,
  "safety_gate_pass": true
}

otto_set_burnout

Manually set burnout level.

Input:

{
  "level": "yellow"
}

otto_set_energy

Manually set energy level.

Input:

{
  "level": "low"
}

Safety Gating

The MCP server enforces Otto's safety invariants:

State	Max Thinking Depth
energy=depleted	minimal
energy=low	standard
burnout>=ORANGE	standard
burnout=RED	minimal
energy=high	ultradeep (if requested)

Rule: Safety state ALWAYS overrides requested depth. Can reduce, never increase.

Use Cases

1. Cross-tool safety: Check cognitive state before starting complex operations in any tool
2. Context awareness: Let AI assistants know your current capacity
3. Intervention routing: Route messages to appropriate experts based on emotional signals
4. Session calibration: Set focus/urgency at the start of work sessions

Development

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

# Run server directly
python -m otto_mcp.server

License

MIT License - see LICENSE for details.

References

• MCP Specification
• Otto
• ThinkingMachines batch-invariance [He2025]