Model Context Protocol (MCP)

1. The Problem: The "API Glue" Trap

Traditionally, connecting an AI to a tool (like your Calendar or Zotero) required writing custom code for every single interaction. This "glue code" is fragile and unscalable.

• The Consequence: Agents remain isolated in the browser, unable to "touch" your local files or specific specialized databases.

2. The Solution: MCP Architecture

MCP (Model Context Protocol) is an open standard that abstracts the connection.

• The Client: The AI (Claude, Gemini, Cursor).
• The Host: The Application (Docker, VS Code).
• The Server: The Connector (Obsidian, Brave Search, Google Drive).

Metaphor: Instead of soldering a different wire for every device (API Glue), we just plug in a standardized cable (MCP).

3. Practical Implementation: The Docker Gateway

We use Docker to manage these connections locally, keeping your data secure.

Step 1: The Config

We configure an mcp_config.json file to tell our AI where to look.

{
  "mcpServers": {
    "obsidian": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "mcp/obsidian"]
    },
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"]
    }
  }
}

Step 2: The Agentic Workflow

Once connected, you simply prompt naturally:

"Search my Obsidian vault for 'GeoAI' and verify any external claims using Brave Search."

The AI automatically selects the correct tool (Obsidian MCP) and the correct verifier (Brave MCP) without you writing a single line of Python.

4. Advanced: Custom MCP Servers

For specialized research specific to Dr. Sounny's work, we can build custom servers.

• Example: An MCP Server that connects directly to the ISU Library Database.
• Result: Your local AI agent gives you citations available physically in your specific library.