Model Context Protocol
Is a standard to integrate LLM with external systems. Shift the burden of tool definition & execution into a MCP Server. Different from API, as schema & functions are defined.
Terminology¶
- Tools
- Available to the AI to do/interact with external things
- With Schema and Functions
- Prompts, which are similar to ~Claude Code
- Ressources to expose data to client and fetch information

Notes¶
- Overview
Communication¶
- Transport agnostic
- MCP spec define messages
- Types
- HTTP, WebSockets
- Hosted by service provider
- STDIO
- Local process
- HTTP, WebSockets
Popular ones¶
- Playwright for web automation
Development¶
MCP Server¶
- MCP Inspector can be called with
mcp dev mcp_server.py, with the last bit - Expose Tools, Ressources
Ressources¶
- Overview
- Type of ressources
- Direct Ressources
- Templated Ressources

- Example
- Autocomplete when the user uses
@file.ext - Pull up the content of the document when the prompt refer to said document
- Autocomplete when the user uses
Prompts¶
- Tailored to the purpose of the MCP server
- Example
- User ask to format a document in Markdown
- User will initiate this with
/and specify with ID
MCP Client¶
- Overview

- Need some cleanup code



