MCP Server for Protocol Specifications and AI Test Generation

[amiable-dev]

MCP Server for Protocol Specifications

Overview

Create an MCP (Model Context Protocol) server that exposes Conductor's protocol specifications and enables AI-powered test generation. This infrastructure enables hardware-free testing and AI-assisted development.

Motivation

• Enable AI/LLMs to understand Conductor's protocols (MIDI, Gamepad, OSC)
• Generate realistic test data without physical hardware
• Provide protocol specs as structured resources
• Foundation for AI-assisted debugging and test creation

Architecture

New Crate: conductor-mcp-server

conductor-mcp-server/
├── Cargo.toml
├── src/
│   ├── main.rs              // MCP server binary
│   ├── server.rs            // MCP protocol implementation
│   ├── resources.rs         // Protocol spec resources
│   ├── tools.rs             // Test generation tools
│   └── schema/
│       ├── midi_spec.json       // MIDI protocol spec
│       ├── gamepad_spec.json    // Gamepad protocol spec (v3.0)
│       ├── input_event.json     // InputEvent enum structure
│       └── action_spec.json     // Action types spec

MCP Resources

Resources exposed via conductor:// URI scheme:

Protocol Specifications

conductor://protocols/midi

{
  "name": "MIDI Protocol",
  "events": {
    "NoteOn": { "note": "0-127", "velocity": "0-127" },
    "NoteOff": { "note": "0-127", "velocity": "0-127" },
    "CC": { "controller": "0-127", "value": "0-127" },
    "PitchBend": { "value": "0-16383" }
  },
  "id_range": "0-127"
}

conductor://protocols/gamepad

{
  "name": "Gamepad Protocol (v3.0)",
  "buttons": {
    "range": "128-144",
    "mapping": {
      "128": "South (A/Cross/B)",
      "129": "East (B/Circle/A)",
      ...
    }
  },
  "axes": {
    "range": "128-133",
    "analog_sticks": ["128-131"],
    "triggers": ["132-133"]
  },
  "id_range": "128-255"
}

conductor://protocols/input-events

• Complete InputEvent enum definition
• Event processing pipeline documentation
• Trigger type mappings

conductor://recordings/{name}

• Captured device recordings (JSON format)
• Stored in .conductor/recordings/
• Used for replay and AI training

MCP Tools

Tools available to AI/LLMs for test generation:

generate_test_sequence

Generate realistic input sequences based on natural language prompts.

Input:

{
  "prompt": "Generate a sequence simulating FPS gameplay with WASD movement",
  "protocol": "gamepad",
  "duration_ms": 5000,
  "output_format": "json"
}

Output:

{
  "sequence": [
    { "timestamp": 0, "event": "ButtonPressed", "button": 131, "comment": "W key (forward)" },
    { "timestamp": 100, "event": "AxisChanged", "axis": 130, "value": 255, "comment": "Right stick X (look right)" },
    ...
  ]
}

validate_message

Validate protocol message correctness.

Input:

{
  "protocol": "gamepad",
  "event": { "type": "ButtonPressed", "button": 128 }
}

Output:

{
  "valid": true,
  "warnings": [],
  "suggestions": ["Button 128 is South button (A/Cross/B)"]
}

suggest_test_cases

AI suggests edge cases and test scenarios.

Input:

{
  "feature": "VelocityRange trigger",
  "protocol": "midi"
}

Output:

{
  "test_cases": [
    "Boundary: velocity 40 (soft/medium boundary)",
    "Boundary: velocity 80 (medium/hard boundary)",
    "Edge: velocity 0 (note off)",
    "Edge: velocity 127 (maximum)",
    "Timing: rapid velocity changes"
  ]
}

Implementation Plan

Phase 1: MCP Server Core (2 hours)

• Create conductor-mcp-server crate
• Implement basic MCP protocol handling
• Add JSON schema definitions for protocols
• Implement resource serving

Phase 2: Protocol Specifications (1 hour)

• Extract MIDI spec from conductor-core
• Extract Gamepad spec (v3.0 mappings)
• Document InputEvent structure
• Create JSON schemas

Phase 3: AI Tools Implementation (2 hours)

• Implement generate_test_sequence tool
• Implement validate_message tool
• Implement suggest_test_cases tool
• Add LLM prompt templates

Phase 4: Testing & Documentation (1 hour)

• Test with Claude Desktop MCP integration
• Test with other MCP clients
• Write usage documentation
• Add examples

Usage Examples

Running the MCP Server

# Start MCP server
conductor-mcp-server --port 3000

# Or via stdio for Claude Desktop integration
conductor-mcp-server --stdio

Claude Desktop Configuration

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "conductor": {
      "command": "conductor-mcp-server",
      "args": ["--stdio"]
    }
  }
}

AI-Powered Test Generation

User: Generate a test sequence for a fighting game combo
AI: [Uses generate_test_sequence tool]
    - Button 128 (A) press at 0ms
    - Button 129 (B) press at 100ms  
    - Button chord [130, 131] at 200ms
    - Creates realistic frame-perfect timing

Benefits

1. Hardware-Free Testing: Generate test data without physical devices
2. AI-Assisted Development: Natural language → test sequences
3. Protocol Documentation: Machine-readable specs
4. Test Coverage: AI suggests edge cases
5. Future Protocols: Easy to add OSC, keyboard, etc.

Dependencies

[dependencies]
serde = { workspace = true }
serde_json = { workspace = true }
tokio = { workspace = true }
conductor-core = { path = "../conductor-core" }
# MCP SDK (to be determined)

Timeline

Total: 6-8 hours

• Phase 1: 2 hours (core server)
• Phase 2: 1 hour (protocol specs)
• Phase 3: 2 hours (AI tools)
• Phase 4: 1 hour (testing/docs)

Blockers

None - this is foundational work that doesn't depend on other issues.

Blocked Issues

• AI-Powered Multi-Protocol Message Generator #3: AI-Powered Message Generator (depends on MCP server)
• macOS BLE Gamepad Support: Xbox Controller not detected via gilrs #1: Multi-backend gamepad support (benefits from test infrastructure)

Labels

enhancement, testing, ai, infrastructure, v3.1

[amiable-dev]

Strategic Pivot: Superseded by #5

After further analysis, we've decided on a more ambitious and valuable approach: building a crowdsourced pattern platform instead of synthetic AI generation.

Superseded by: #5 (Crowdsourced Pattern Platform)

Why the Change?

Original approach (this issue):

• AI generates test data from protocol specs
• Synthetic patterns based on rules
• Limited realism

New approach (#5):

• Capture real human input patterns
• Crowdsource from actual users
• AI trains on real data
• Much more realistic test scenarios

MCP Integration Still Relevant

The MCP server concept from this issue is incorporated into #5:

• Cloud API exposes patterns as MCP resources
• AI tools query real pattern database
• Local MCP server for offline use

Status

Closing this issue in favor of the more comprehensive solution in #5.