conductor-mcp-server

bensleveritt/conductor-mcp-server

3.2

If you are the rightful owner of conductor-mcp-server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

Conductor MCP Server is an orchestration server for AI-assisted development workflows, integrating Ollama models using Bun and TypeScript.

Tools
9
Resources
0
Prompts
0

Conductor MCP Server šŸŽ¼

An MCP (Model Context Protocol) server that conducts and orchestrates Ollama models for AI-assisted development workflows. Built with Bun and TypeScript.

Features

  • šŸ¦™ Ollama Integration: Seamless connection to local Ollama models
  • šŸ”„ Conversation Continuity: Maintain context across tool calls with continuation IDs
  • šŸ§  Multi-Step Workflows: Debug, plan, and analyze with iterative reasoning
  • šŸŽ­ Multi-Model Consensus: Consult multiple models for complex decisions
  • šŸŽÆ Thinking Modes: Adjustable reasoning depth (minimal ā†’ max)
  • šŸ› ļø Rich Tool Set: From simple chat to complex code review workflows

Prerequisites

  • Bun v1.2+
  • Ollama running locally
  • At least one Ollama model installed (e.g., ollama pull qwen2.5:latest)

Installation

# Clone the repository
git clone <your-repo-url>
cd conductor-mcp-server

# Install dependencies
bun install

Configuration

Create a .env file in the root directory (optional):

# Ollama server URL
OLLAMA_BASE_URL=http://localhost:11434

# Default model to use
DEFAULT_MODEL=qwen2.5:latest

# Comma-separated list of tools to disable
DISABLED_TOOLS=codereview,precommit

# Maximum conversations to keep in memory
MAX_CONVERSATIONS=100

Usage

Running the Server

# Development mode (with watch)
bun run dev

# Production mode
bun run start

Configuring in Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "conductor": {
      "command": "bun",
      "args": ["run", "/path/to/conductor-mcp-server/index.ts"]
    }
  }
}

Or with environment variables:

{
  "mcpServers": {
    "conductor": {
      "command": "bun",
      "args": ["run", "/path/to/conductor-mcp-server/index.ts"],
      "env": {
        "OLLAMA_BASE_URL": "http://localhost:11434",
        "DEFAULT_MODEL": "qwen2.5:latest"
      }
    }
  }
}

Available Tools

Core Tools

chat

General conversation and collaborative thinking. Use for brainstorming, getting second opinions, and exploring ideas.

Parameters:

  • prompt (required): Your question or idea
  • files (optional): File paths for context
  • model (optional): Model to use (default: qwen2.5:latest)
  • temperature (optional): 0-1, controls creativity
  • thinking_mode (optional): minimal, low, medium, high, max
  • continuation_id (optional): Continue previous conversation
listmodels

List all available Ollama models with their details.

version

Display server version, configuration, and available tools.

Workflow Tools (Priority 2)

debug

Systematic debugging with multi-step hypothesis testing. Guides through root cause analysis with evidence collection.

Key Features:

  • Step-by-step investigation
  • Hypothesis tracking with confidence levels
  • File examination tracking
  • Backtracking support
thinkdeep

Extended reasoning for complex analysis. Use for architecture decisions, performance challenges, and deep problem exploration.

Thinking Modes:

  • minimal: Quick, focused analysis
  • low: Basic reasoning
  • medium: Moderate analysis
  • high: Deep analysis with multiple perspectives
  • max: Comprehensive exploration
planner

Interactive project planning with revision and branching capabilities. Breaks down complex tasks into manageable steps.

Features:

  • Sequential planning
  • Step revision
  • Alternative branch exploration
  • Dependency tracking
consensus

Multi-model consultation for complex decisions. Gather perspectives from different models with configurable stances.

Parameters:

  • models: Array of models with optional stances (for/against/neutral)
  • Systematic consultation process
  • Final synthesis of all perspectives

Specialized Tools (Priority 3)

codereview

Comprehensive code review covering quality, security, performance, and architecture.

Review Types:

  • full: Complete analysis
  • security: Security-focused
  • performance: Performance-focused
  • quick: Critical issues only
precommit

Pre-commit validation to prevent bugs and quality issues from being committed.

Features:

  • Staged/unstaged change analysis
  • Security vulnerability detection
  • Breaking change identification
  • Commit message suggestions

Conversation Continuity

All workflow tools support continuation_id for maintaining context:

// First call creates a conversation
chat({ prompt: "Explain async/await" })
// Returns: conversation_id: "conv_123..."

// Continue the conversation
chat({
  prompt: "Now show an example",
  continuation_id: "conv_123..."
})

Multi-Step Workflows

Workflow tools (debug, thinkdeep, planner, consensus) use a step-based approach:

debug({
  step: "Initial investigation of memory leak",
  step_number: 1,
  total_steps: 3,
  next_step_required: true,
  findings: "High memory usage in production",
  hypothesis: "Possible event listener leak",
  confidence: "medium"
})

Example: Multi-Model Consensus

consensus({
  step: "Should we migrate to microservices?",
  step_number: 1,
  total_steps: 4, // Initial + 2 models + synthesis
  next_step_required: true,
  findings: "Current monolith has scaling issues",
  models: [
    { model: "qwen2.5:latest", stance: "for" },
    { model: "llama3.1:latest", stance: "against" }
  ]
})

Development

# Run in development mode
bun run dev

# Build for distribution
bun run build

# Run tests (when added)
bun test

Project Structure

conductor-mcp-server/
ā”œā”€ā”€ src/
ā”‚   ā”œā”€ā”€ tools/          # Tool implementations
ā”‚   ā”‚   ā”œā”€ā”€ chat.ts
ā”‚   ā”‚   ā”œā”€ā”€ debug.ts
ā”‚   ā”‚   ā”œā”€ā”€ thinkdeep.ts
ā”‚   ā”‚   ā””ā”€ā”€ ...
ā”‚   ā”œā”€ā”€ types/          # TypeScript types and schemas
ā”‚   ā”‚   ā””ā”€ā”€ index.ts
ā”‚   ā””ā”€ā”€ utils/          # Utilities
ā”‚       ā”œā”€ā”€ ollama-client.ts
ā”‚       ā”œā”€ā”€ memory.ts
ā”‚       ā””ā”€ā”€ config.ts
ā”œā”€ā”€ index.ts            # Main server entry point
ā”œā”€ā”€ package.json
ā””ā”€ā”€ tsconfig.json

Architecture

  • MCP Server: Built on @modelcontextprotocol/sdk
  • Ollama Client: REST API integration with streaming support
  • Memory Management: Conversation history with automatic cleanup
  • Tool Registry: Dynamic tool loading with disable capability
  • Zod Validation: Type-safe input validation

Troubleshooting

Ollama Connection Issues

# Check if Ollama is running
curl http://localhost:11434/api/tags

# Start Ollama (if not running)
ollama serve

Model Not Found

# List available models
ollama list

# Pull a model
ollama pull qwen2.5:latest

MCP Server Not Responding

  1. Check Claude Desktop logs
  2. Verify the server path in config
  3. Ensure Bun is in your PATH
  4. Try running the server manually: bun run index.ts

Contributing

Contributions are welcome! Please feel free to submit issues and pull requests.

License

MIT

Credits

Inspired by zen-mcp-server