Build Your First MCP Tool

AI News Search and Newsletter in Python

There is a quiet moment in many people’s “AI journey” where chat windows stop being enough.

You’re no longer impressed by “Explain transformers in simple terms”.
You want something more practical:

• “Look up this week’s AI news and summarise it for my blog readers.”
• “Draft a newsletter based on real headlines, not hallucinated ones.”
• “Help me keep my weekly content consistent, even when I’m tired.”

MCP tools — small, purpose-built functions an AI agent can call directly — are where this need is answered.

In this post, we’ll build a small but complete example:

An MCP-style Python service that:

1. searches the web for AI-related news, and
2. uses a local model (via Ollama) to draft an AI newsletter.

We’ll keep the code simple and the explanations gentle.
You do not need to be a backend expert for this.

---

What You’re Building: An MCP-Style FastAPI News and Newsletter Tool

We’ll create a tiny HTTP service using FastAPI with two main endpoints:

1. /search_ai_news
   – Takes a topic and a time window (e.g. “AI”, “last 3 days”).
   – Queries a news API to fetch recent AI headlines and summaries.

2. /newsletter
   – Uses the search results.
   – Calls a local LLM via Ollama.
   – Returns a warm, human-readable newsletter draft aimed at your readers.

This FastAPI service can:

• be called directly (with curl or Python),
• be wrapped as a genuine MCP tool once you add an MCP server on top (more on that in Step 7),
• and slot into your Obsidian + GitHub Pages workflow.

Think of it as your first “practical AI helper” that actually does work for your blog.

A quick note on naming: the Model Context Protocol (MCP) is an open protocol that lets AI applications discover and call external tools through a standard interface. What we’re building here is “MCP-style” in spirit — a small, well-defined tool boundary — but it is a plain FastAPI REST service, not an actual MCP server. Real MCP servers speak JSON-RPC over stdio or Streamable HTTP, as defined in the MCP specification. Step 7 shows exactly where the gap is and how to close it.

---

Prerequisites: Python, Ollama, and a News API Key

You’ll need:

• Basic Python knowledge (functions, virtualenv, running scripts).
• Python 3.10+ recommended.
• A terminal that doesn’t scare you completely.
• A news API key. In this example, we’ll assume something like NewsAPI or a similar REST news provider. (You can adapt to Tavily, SerpAPI, etc.)
• Ollama installed and running locally.

• A model pulled in Ollama, for example:

  ollama pull llama3
  

We’ll call the model "llama3" in the code, but you can change it.

---

Step 1 — Project layout

🔒 Subscribe to keep reading.

Step 2 — Configuration with environment variables

🔒 Subscribe to keep reading.

Step 3 — A tiny client for the news API

🔒 Subscribe to keep reading.

Step 4 — A tiny client for Ollama (local LLM)

🔒 Subscribe to keep reading.

Step 5 — Wiring everything with FastAPI

🔒 Subscribe to keep reading.

Step 6 — Using it directly (without MCP yet)

🔒 Subscribe to keep reading.

Step 7 — Where MCP comes in

🔒 Subscribe to keep reading.

Step 8 — Example: calling the tool from a small “agent” script

🔒 Subscribe to keep reading.

Step 9 — Integrating the MCP Tool with an Obsidian and Jekyll Blogging Workflow

🔒 Subscribe to keep reading.

Key Takeaways: Building Your First MCP Tool

🔒 Subscribe to keep reading.

References

🔒 Subscribe to keep reading.

Practical Upgrade Path

🔒 Subscribe to keep reading.