Bridge between Claude Desktop and Claude Code via MCP + Agent SDK.

Conductor is an MCP server that lets Claude Desktop execute tasks using Claude Code's capabilities. Send vision documents, prompts, or instructions from Desktop and watch them executed with full file access, terminal commands, and code editing.

Claude Desktop  ──MCP──►  Conductor  ──Agent SDK──►  Claude Code
     ▲                                                    │
     └────────────────── results ─────────────────────────┘

1. Claude Desktop connects to Conductor via MCP (stdio)
2. Desktop calls Conductor tools with prompts/vision docs
3. Conductor uses the Agent SDK to execute with Claude Code's capabilities
4. Results stream back to Desktop

cd conductor
npm install

cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY

Get your API key from console.anthropic.com.

npm run build

Add Conductor to your Claude Desktop MCP configuration (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "conductor": {
      "command": "node",
      "args": ["/Users/YOUR_USERNAME/Documents/conductor/dist/index.js"],
      "env": {
        "ANTHROPIC_API_KEY": "your-api-key-here"
      }
    }
  }
}

Replace YOUR_USERNAME with your actual username and add your API key.

Quit and reopen Claude Desktop to load the new MCP server.

Set the working directory for subsequent executions.

set_project({ path: "/Users/you/projects/myapp" })

Execute any prompt using Claude Code's full capabilities.

execute({
  prompt: "Find all TODO comments and create a summary",
  project: "/path/to/project",  // optional if set_project was called
  allowedTools: ["Read", "Glob", "Grep"]  // optional, defaults to all
})

Execute a vision/planning document. Designed for handoffs from planning sessions.

execute_vision({
  vision: "# Feature: User Authentication\n\n...",
  flowcharts: "```mermaid\n...\n```",  // optional
  project: "/path/to/project",
  instruction: "start with tests"  // optional
})

Continue a previous session with follow-up instructions.

continue_session({
  prompt: "Now add error handling to what you just built"
})

Get current Conductor state (project, session info).

In Claude Desktop:

"Use conductor to set the project to ~/projects/myapp"

"Use conductor to execute: analyze the codebase structure and identify the main entry points"

"Use conductor to execute_vision with this vision document: [paste your VISION.md content]"

"Use conductor to continue_session: now write tests for what you just implemented"

Conductor runs with bypassPermissions mode, meaning Claude Code can read, write, and execute commands without prompts. This is intentional for seamless automation but means you should:

• Only use on trusted projects
• Review what you're asking it to execute
• Consider the implications of file/command access

# Run in development mode
npm run dev

# Type check
npm run typecheck

# Build for production
npm run build

• Agent SDK: 0.2.4
• MCP SDK: 1.25.2