AI Blog: Building Your Own Model Context Protocol (MCP)

As discussed in our earlier blog, a Model Context Protocol (MCP) acts as a universal adaptor, streamlining the interaction between AI agents and various tools, data sources and services.  Now, let’s take a leap forward and explore how to build a real-world MCP: an AI Meeting Summariser.

Crafting Your Own MCP: The AI Meeting Note Summariser

This application is designed to transform long meeting transcripts to concise and actionable key points.  It's a good example of how MCP can simplify complex integrations and enhance the capabilities of your AI agents.  

The Core Logic: How Our Summariser Works

Large Language Models (LLMs) enable MCP to perform its magic.  Here's a breakdown of its components and workflow:

• Model: the brain of our summariser is a powerful LLM, such as OpenAI's ChatGPT, Google's Gemini or Anthropic's Claude.  These models will play the role of summarising extensive text into clear key points
• Role: the LLM's role is pre-defined to transform thousand words of meeting transcripts into concise summaries, highlighting the most critical information and decisions
• Client (User Interface): The user interacts with a simple web application.  This front-end, built with basic HTML and JavaScript, provides an intuitive interface for users to paste or upload their meeting notes.  Users can either upload .txt or .docx files containing their meeting transcripts or simply paste the text directly to the input box
• Process (Backend Logic / API): once the meeting notes are submitted through the MCP client, the text is securely transmitted to the chosen LLM API (e.g. OpenAI API) via a RESTful API call.  The LLM processes the input, generates the summary and sends it back to the client for display.  This backend interaction sounds quite straightforward is where the MCP truly make a different by standardising and simplifying the communication between the application and the LLM service.

Here is a part of the code we have written for the AI Meeting Summariser.

Even if you can’t write a single line of code, you can still build powerful AI tools by thinking like a prompt designer.  You define how the AI should behave (context), what model to use and what input it will receive (prompt).  Then, tools like ChatGPT, Gemini or automation platforms like Zapier can help you build and run your AI agent, no coding required.

Join us next time for more exciting developments in the world of AI MCPs!