Experienced Developer Path
You know your way around systems. This path skips the hand-holding and goes straight to protocol internals, full implementations, parallel tool calls, error handling, and building production-ready MCP servers.
What you'll come away with
- ✓ The full tool_use / tool_result loop including parallel calls and stop_reason mechanics
- ✓ JSON-RPC internals — the actual wire format of MCP messages
- ✓ Building a working MCP server in Python and TypeScript
- ✓ Transport selection: stdio vs HTTP+SSE and the new Streamable HTTP spec
- ✓ Error handling at both the protocol and tool layer
Your curriculum
Architecture Overview — For Experienced Developers
How tools, skills, and MCP stack together — the three extensibility primitives, the host routing layer, and the mental model that ties them all.
What is an LLM — For Experienced Developers
The transformer architecture, tokenization mechanics, and inference pipeline — everything you need to understand how LLMs actually work under the hood.
Tools: A Deep Dive
What tools actually are, how the request/execute/return loop works, parallel calls, error handling, and how to write tool definitions that the model uses correctly.
MCP: Model Context Protocol
The open protocol that standardizes how AI agents connect to external systems. JSON-RPC internals, transports, the three primitives, and how to build a custom server.
Skills — For Experienced Developers
Just-in-time retrieval-augmented prompting — how skill files work, how to structure them, and how they compare to RAG and fine-tuning.