LOCO VS Code Extension (MCP-ready)

This extension demonstrates a low-code AI workflow in VS Code:

• Generate Python/C++ stubs from natural language (local mock provider)
• Inline generation in editor via trigger like lc:
  • Tip: type lc: class MyNode base=myframework.base.NodeBase params=threshold:float=0.5:阈值 to inline-generate a class
• Built-in HTTP endpoint for demo (/generate)
• Embedded MCP stdio server for tools like Cursor

No external MCP server is required anymore. The previous mcp-server/ folder has been removed; everything is embedded in the extension.

Structure

• Root — VS Code extension (TypeScript). Build with npm run compile.

Quick start (Windows PowerShell)

1. Install dependencies and run tests:

cd "d:\1.speedcode\gitlab\kunwu\loco_vscode_extension_mcp"; npm install; npm run test

1. Press F5 in VS Code to launch the Extension Development Host.

2. Try these commands in the Command Palette:

• "LOCO: 生成拓展代码" — Prompt for a description and open generated code
• "LOCO: 使用选区生成并插入" — Use selected text as prompt and insert generated code
• "LOCO: 切换 Inline 设置" — Toggle inline generation and backend (local/MCP/Tongyi)
• "LOCO: 复制 Cursor stdio 配置（绝对路径）" — Copy a ready-to-paste MCP stdio config with an absolute path to the embedded server

1. Inline generation: In any editor, type a trigger like # lc: parse JSON (or // lc parse args) and accept the suggestion.

• Class intent: start with lc: class ... (or include 类) and optional keys: base=..., params=..., lang=python|cpp.
• If loco.inline.generateClass is enabled (default true), descriptions without explicit def/function will prefer generating a class.

Built-in services

• HTTP demo server

  • Automatically starts (by default) on port 8787 at activation
  • Endpoint: POST <http://127.0.0.1:8787/generate> with JSON body { description, language }
  • Configure via Settings: loco.mcp.enableBuiltInServer and loco.mcp.port

• MCP stdio server (embedded)

  • Exposes tool generate(description, language?)
  • Use "LOCO: 复制 Cursor stdio 配置（绝对路径）" to copy a cross-platform config snippet for Cursor

Settings

• loco.provider — Default model provider label (mocked)
• loco.language — Default target language: python | cpp
• loco.inline.enabled — Enable/disable inline generation
• loco.inline.trigger — Trigger text, default lc:
• loco.inline.backend — Use local mock, mcp (HTTP), or tongyi
• loco.inline.generateClass — Prefer generating a class for inline triggers (default true)
• loco.tongyi.model / loco.tongyi.apiBase — Tongyi model and optional base URL
  • Use command "LOCO: 设置通义千问 API Key" to store your API key securely
• loco.mcp.baseUrl — Base URL for HTTP backend (default http://localhost:8787)
• loco.mcp.enableBuiltInServer — Start embedded HTTP server
• loco.mcp.port — HTTP server port

Notes

• This is a demo using a local mock generator. You can wire real providers (e.g., Tongyi/QWen/OpenAI) inside src/generator.ts.
• For Cursor MCP stdio mode, the copied config runs node <absolute-path-to>/mcpStdioServer.js directly and works on Windows/Mac/Linux.

Cursor-only distribution (single-file executable)

If you want to ship a standalone MCP server for Cursor users (no Node/VS Code required), you can build single-file executables using pkg:

1. Build binaries (from project root):

cd "d:\1.speedcode\gitlab\kunwu\loco_vscode_extension_mcp"; npm install; npm run build:stdio:exe

Artifacts will be output to ../release/ as:

• Windows: loco-mcp-stdio-win.exe
• macOS: loco-mcp-stdio-macos
• Linux: loco-mcp-stdio-linux

1. Cursor config example (Windows):

{
  "mcpServers": {
    "loco-lowcode": {
  "command": "C:/Path/To/release/loco-mcp-stdio-win.exe",
      "args": [],
      "disabled": false,
      "description": "LOCO 单文件 MCP stdio 服务器"
    }
  }
}

1. macOS/Linux users: point command to the corresponding binary in release/.

Alternatively, publish as an npm CLI (bin pointing to dist/mcpStdioServer.js) and let users set command: "loco-mcp-stdio" with empty args.

Publish to VS Marketplace (brief)

1. Prereqs: Create a Publisher, install vsce.

2. Ensure package has required fields (name, displayName, publisher, version, engines.vscode, icon).

3. Build and package:

cd "d:\1.speedcode\gitlab\kunwu\loco_vscode_extension_mcp"; npm run compile; npx vsce package

1. Publish:

npx vsce publish