Next-Generation MCP Server

A sandboxed file-operations MCP server that showcases the advanced Context features — logging, progress reporting, and user elicitation — with two ways to drive it: a Claude CLI (client.py) and a Groq web chat (app.py).

Capabilities

🛠️ Tools (model-controlled — the agent calls them):

📄 Resources (app-controlled): file:///{file_name} (read file) · dir://. (list dir) · stats://project (project statistics)

💬 Prompts (user-controlled): code_review · documentation_generator (elicitation) · generate_tests · explain_code

Related MCP server: Sparkle MCP Server

Project structure

next-generation-mcp-server/
├── fileops/                # modular MCP server package
│   ├── config.py           # BASE_DIR, server name/instructions, ignore list
│   ├── paths.py            # get_path() path-traversal guard
│   ├── schemas.py          # Pydantic models (elicitation)
│   ├── server.py           # builds `mcp`, registers tools/resources/prompts
│   ├── tools.py            # @mcp.tool definitions
│   ├── resources.py        # @mcp.resource definitions
│   └── prompts.py          # @mcp.prompt definitions
├── server.py               # entry point  →  python server.py
├── client.py               # Claude CLI client (Anthropic)
├── app.py                  # Groq chat UI (Chainlit)
├── chainlit.md
├── tests/test_server.py    # smoke test of all capabilities
├── pyproject.toml · requirements.txt · .env.example · .gitignore · README.md

Setup

cd next-generation-mcp-server
python -m venv .venv
.\.venv\Scripts\Activate.ps1        # Windows  (macOS/Linux: source .venv/bin/activate)
pip install -r requirements.txt

Add a key to .env (Copy-Item .env.example .env):

• GROQ_API_KEY — free at console.groq.com, used by the web UI (app.py).

• ANTHROPIC_API_KEY — used by the CLI (client.py).

Run

Web chat UI (Groq):

chainlit run app.py -w          # → http://localhost:8200

Chat to use the file tools; slash commands for the rest: /dir · /read <f> · /stats · /tools · /review <f> · /tests <f> · /explain <f> · /docs · /help. You can also attach a file and say "review this file".

Claude CLI (Anthropic):

python client.py server.py

Run the server directly / tests:

python server.py
python tests\test_server.py     # or: python -m pytest

How it works

• The server runs over STDIO; the client/UI spawn it as a subprocess.

• Tools are model-controlled (the agent calls them in its loop).

• Resources are app-controlled (/dir, /read, /stats).

• Prompts are user-controlled (/review, /tests, /explain, /docs).

Based on the IBM Skills Network "Enhanced MCP Server" lab, refactored into a modular package and extended with more tools, resources, prompts, and a Groq UI.