MCP Implementation (Phase 1 — local test)

A minimal MCP server (the "skills library") with one sample skill search_news. Goal: prove that local Claude Code can auto-call a skill over MCP.

Folder

mcp implementation/
├─ server/
│  └─ main.py          # MCP server + the search_news skill
├─ requirements.txt    # dependency: mcp
├─ .mcp.json           # tells Claude Code to connect to the local server
└─ README.md           # this file

Related MCP server: MCP Practice

How to run

1. (Recommended) create a virtual environment:

   python -m venv .venv
   .venv\Scripts\activate          # Windows PowerShell

2. Install the dependency:

   pip install -r requirements.txt

3. Start the server (leave this terminal running):

   python server/main.py

   It serves at: http://127.0.0.1:8000/mcp

4. Open Claude Code in this folder (mcp implementation) so it reads .mcp.json. Approve the new MCP server if prompted. Check it connected with /mcp.

How to verify success

In Claude Code, type:

find today's news about AI

Claude should automatically call the search_news skill and reply with:

Today's news about 'AI': [demo result] The LLX Agent MCP server is working!

✅ If you see that, the full chain works: local Claude → auto MCP call → skill runs → returns result.

What changes when we move to the cloud

Nothing in the skill logic. We deploy server/main.py to Azure Container Apps, and the only edit on the employee side is the url in .mcp.json:

"url": "https://<your-azure-app>.azurecontainerapps.io/mcp"

(plus an auth token header, added in the security phase).