AI Agents Guide: MCP, Claude Code and Local LLMs
The curated learning path: from the MCP concept through Claude Code to a local LLM
This guide is a through-line across the topic of AI agents, not a single tutorial. Each stop links the matching article in detail. The order is a recommendation – if you know the basics, jump straight to practice or self-hosting.
This blog is itself run by an AI agent via its own MCP server – so the examples here aren't theoretical, they're what runs every day.
1. Basics
First understand the concept: what is an agent, and how does it get tools?
- Understanding LLM agents: from chatbot to agent – the agent loop without hype.
- What is an MCP server? – how an LLM gets tools in the first place.
2. Tools & practice
From your first agent in the terminal to a permanently running server agent.
- Claude Code for beginners – set up an AI agent in your terminal.
- AI in the terminal: CLI agents overview – the why behind the workflow.
- Claude Code with Remote Control – the agent that lives permanently on your server.
3. Self-host & build
Run your own models and build your own tools.
- Local LLMs with Ollama – self-host language models without the cloud.
- LibreChat with Claude & Gemini – a UI for the MCP ecosystem.
- Developing MCP servers in Python – from empty repo to the first tool.
- MCP server doesn't survive a deploy – SSE vs. Streamable-HTTP, the transport pitfall.
An agent setup like this barely needs hardware – my MCP server and the local models run on a power-efficient mini PC that also carries the rest of the homelab:
Ad · Affiliate link – if you buy through it, I may earn a commission. It doesn’t change the price for you.
Related topics
- All Software & Web articles – the full overview.
- Homelab Self-Hosting Guide – the foundation your agents run on.