learn.ai

Week 3 of 5

MCP, LangGraph, memory and multi-agent

Expose your tools over MCP, rebuild the agent as a LangGraph graph with memory and human approval, and try an orchestrator with parallel workers.

0%

Pick a start date to see real dates.

0%
Today

Build this week

  • Tools and retrieval exposed as an MCP server, used from an MCP client
  • The agent rebuilt in LangGraph with a checkpointer and a long-term store
  • Human approval before risky tools
  • An orchestrator that splits a query into parallel workers and merges the results
  1. Day 151 h 30 min

    MCP basics

    Notes
  2. Day 161 h 30 min

    Your MCP server

    Notes
  3. Day 171 h 30 min

    LangGraph state

    Notes
  4. Day 181 h 30 min

    Persistence and approvals

    Notes
  5. Day 194 h 30 min · weekend

    Agent rebuild

    Notes
  6. Day 203 h 50 min · weekend

    Many agents

    Notes
  7. Day 211 h 20 min

    Week review

    • 50 min
    • 30 min
    Notes

If you have time

Checkpoint

Answer out loud, then rate yourself.

MCP or plain function calling? When does each make sense?
When is a multi-agent design worth its cost?
What types of memory does an agent have, and where does each live?