gpt-5-plan MCP Server (npx-first)

Generate clear development plans using GPT-5 with a structured Japanese YAML specification. This server runs via npx, so you don't need to clone or build locally.

Key features

• npx-friendly: install-and-run with a single command
• OpenAI-compatible: works with OpenAI or compatible endpoints via OPENAI_BASE_URL
• Safe defaults: STDIO transport for broad MCP client compatibility

Requirements

• Node.js 18+
• An MCP-capable client (Cursor, Claude Code, VS Code MCP, Windsurf, etc.)
• OpenAI API Key (OPENAI_API_KEY)

Getting started (npx, no clone required)

Most MCP clients support launching servers via npx. Use the standard configuration below and inject your environment variables.

Standard client config (JSON)

{
  "mcpServers": {
    "gpt-5-plan": {
      "command": "npx",
      "args": ["gpt-5-plan-mcp@latest"],
      "env": {
        "OPENAI_API_KEY": "sk-REPLACE_ME",
        "OPENAI_MODEL": "gpt-5",
        "OPENAI_TEXT_VERBOSITY": "low"
      },
      "autoStart": true
    }
  }
}

Claude Code (recommended)

Add this server with the CLI (no local clone needed):

claude mcp add gpt-5-plan -s user -e OPENAI_API_KEY="sk-REPLACE_ME" -e OPENAI_MODEL="gpt-5" -e OPENAI_TEXT_VERBOSITY="low" -- npx -y gpt-5-plan-mcp@latest

Remove if needed:

claude mcp remove gpt-5-plan

Cursor

Project-local example in .cursor/mcp.json:

{
  "mcpServers": {
    "gpt-5-plan": {
      "command": "npx",
      "args": ["gpt-5-plan-mcp@latest"],
      "env": {
        "OPENAI_API_KEY": "sk-REPLACE_ME",
        "OPENAI_MODEL": "gpt-5",
        "OPENAI_TEXT_VERBOSITY": "low"
      },
      "autoStart": true
    }
  }
}

VS Code / VS Code Insiders

Use the standard JSON config in your MCP extension, or install via your MCP client UI. Pin a version with @1.0.0 if you need stability.

Other clients (Windsurf, LM Studio, Goose, Qodo Gen, etc.)

• Command: npx
• Args: ["gpt-5-plan-mcp@latest"]
• Env: OPENAI_API_KEY=... and optional variables below

Configuration

• OPENAI_API_KEY (required): your API key
• OPENAI_MODEL (default: gpt-5)
• OPENAI_TEXT_VERBOSITY (low | medium | high, default: low)
• OPENAI_BASE_URL (optional): set when using a compatible proxy endpoint
• OPENAI_REASONING_EFFORT (low | medium | high, default: medium)

Notes:

• The server chooses reasoning effort per request; defaults may be overridden in code paths for gpt5_plan.

Tools

• gpt5_plan: Generate a YAML plan from inputs using the Japanese YAML spec

Input fields (all optional, strings): goal, context, user_request, scope, focus_features, project_type, non_functionals, constraints, kpi_preferences, paneling, panel_count, targets.

Example call (conceptual):

{
  "name": "gpt5_plan",
  "arguments": {
    "user_request": "新規機能Xの開発計画を作って",
    "scope": "full",
    "project_type": "web-app",
    "focus_features": "auth, i18n"
  }
}

Troubleshooting

• 401 Unauthorized: verify OPENAI_API_KEY (no extra whitespace)
• Use Node 18+
• Do not console.log to stdout in STDIO servers; use stderr for logs
• Claude logs: tail -n 50 -f ~/Library/Logs/Claude/mcp*.log

Local development (optional)

You can still clone and build locally if desired:

npm i
npm run build
node build/index.js

License

ISC