Learning Map
1
MCP
MCP
Connect tools, resources and prompts exposed by local or remote servers to an Agent.
2
Context
Context
Inject caller-provided user preferences, roles or permissions as runtime context.
3
State
State
Use AgentState, Command and a checkpointer to save working memory across turns.
4
Delegation
Delegation
Wrap sub-agents as tools so the main Agent can route and coordinate.
5
Data
Data
Use RAG and SQL to let an Agent query documents and structured databases.
Lesson Summary
1
2.1a MCP
Open the explained page
Connect to local and third-party MCP servers and read tools/resources/prompts.
- MCP
- MultiServerMCPClient
- tools
2
2.1b Travel Agent
Open the explained page
Build a flight-search travel Agent with remote Kiwi MCP tools.
- remote MCP
- SSE
- travel tools
3
2.2a Runtime Context
Open the explained page
Read runtime user preferences through context_schema and ToolRuntime.
- context_schema
- ToolRuntime
- preferences
4
2.2b State
Open the explained page
Extend AgentState and use Command(update=...) to read and write cross-turn state.
- AgentState
- Command
- checkpointer
5
2.3 Multi Agent
Open the explained page
Wrap sub-agents as tools for the main Agent to select and call.
- subagent
- delegation
- routing
6
2.4 Wedding Planners
Open the explained page
Coordinate three expert Agents for flights, venues and playlists to plan a wedding.
- coordinator
- state
- specialists
7
Bonus RAG
Open the explained page
Load PDFs, chunk them, create embeddings, run vector retrieval, then wrap retrieval as an Agent tool.
- vector store
- retrieval
8
Bonus SQL
Open the explained page
Wrap SQLDatabase so an Agent can query the Chinook SQLite database.
- SQLDatabase
- query tool
- trace
Completion Criteria
- Explain in your own words what problem each lesson in this module solves.
- Identify the key API in each example and where it fits in the Agent workflow.
- Apply this module's techniques to a new small Agent task.