AI-First Coding

Why bother with AI?

Coding is high on priority list for LLM development

Top models are ranked among top competitive coders in world (o3 - 175th best)

Humans + AI have a superpower advantage over humans without it

AI code tooling progression

Chat - manual implementation

Copilot - working side by side

AI + Tools - orchestration + validation

Current Stack Overview

Claude Desktop most of the time

Some of the time: OpenAI Codex, Claude CLI

MCP-centric approach

MCP Overview

MCP: Model Context Protocol

Open protocol, like USB-C for AI applications

Wide support, including local models, portable tools

Easy to build your own, but rarely a need to

How it works

MCP Hosts: acts as client, connects to server

MCP Clients: hosts maintain connections with servers

MCP Servers: lightweight programs exposing capabilities

Useful coding MCP servers

Filesystem: complete file access (path-restricted)

Git: perform git actions directly, including history

GitHub: pull requests, issues, comments, files

Playwright: browser automation

Postgres: read-only DB access

SQLite: full DB access

Build your own!

Example Configuration


						{
							"mcpServers": {
								"filesystem": {
									"command": "npx",
									"args": [
										"-y",
										"@modelcontextprotocol/server-filesystem",
										"/Users/username/Desktop",
										"/Users/username/Downloads"
									]
								}
							}
						}