MCP.so
ログイン

MCP Lab

@rperezll

基本情報

カテゴリ

AI とエージェント

トランスポート

stdio

作者

rperezll

MCP Lab について

Laboratorio personal con MCP en TypeScript: MCP Server, MCP Client, Claude Desktop y OpenAI API

MCP Lab の設定

標準の設定はありません

このサーバーの README には解析可能な MCP 設定ブロックが含まれていません。インストール手順はリポジトリをご確認ください。

リポジトリ

MCP Lab のツール

ツール一覧はまだありません

この掲載には表示できるツール一覧がまだありません。サーバーにツールがないという意味ではありません。対応機能はプロジェクトのドキュメントをご確認ください。

ツールは README から自動的に抽出されます。メンテナーは ## Tools という見出しの下に記載することで、このタブに反映できます。

概要

What is MCP Lab?

MCP Lab is a personal experimentation repository for learning and testing the Model Context Protocol (MCP) using the TypeScript SDK. It contains multiple sub-projects: a basic MCP server (mcp-server-dummies), a Text-to-SQL server for SQLite (mcp-lab-text2sql), and a client that integrates MCP with OpenAI’s Responses API.

How to use MCP Lab?

Each sub‑project is installed and run separately. Navigate to a project (e.g. mcp-server-dummies) and run npm i, then npm run build (compiles with tsc) and npm run dev (runs with ts-node). Debug with npm run test which launches the MCP Inspector at http://localhost:6274. Servers can be added to Claude Desktop via claude_desktop_config.json under AppData/Roaming/Claude on Windows.

Key features of MCP Lab

  • Basic MCP server for learning and quick testing.
  • Text‑to‑SQL server letting you chat with SQLite databases in natural language.
  • MCP client that bridges MCP tools with OpenAI’s Responses API.
  • Supports debugging via the MCP Inspector visual interface.
  • All projects require Node.js >= v20.
  • Licensed under MIT.

Use cases of MCP Lab

  • Experimenting with the MCP SDK and understanding protocol mechanics.
  • Building and testing a Text‑to‑SQL assistant over a local SQLite database.
  • Connecting MCP servers to an OpenAI‑powered agent (via the provided client).
  • Prototyping and debugging MCP tools before production deployment.

よくある質問

1

What runtime does MCP Lab require?

All projects need Node.js version 20 or higher.

2

How can I debug an MCP server in this lab?

Run `npm run test` inside a server project – this launches the MCP Inspector at `http://localhost:6274` for interactive testing.

3

Can I use these MCP servers with Claude Desktop?

Yes. Edit `claude_desktop_config.json` (located in `AppData/Roaming/Claude` on Windows) and add entries with the `node` command pointing to the compiled `dist/index.js` of each server. Remember to run `npm run build` first.

4

Are these MCP servers directly compatible with OpenAI?

No. MCP tool schemas differ from OpenAI’s function‑calling format. MCP Lab includes a client project (`mcp-client`) that transforms tools for use with the OpenAI Responses API. When defining tool arguments with Zod, OpenAI requires all params be required, forbids `min`/`max` on integers, and disallows `default` values.

5

What data does the Text‑to‑SQL server work with?

The `mcp-lab-text2sql` server interacts with a SQLite database, but the README does not specify which database or its location; the server is designed to chat with your own SQLite data.

コメント

「AI とエージェント」の他のコンテンツ