examples of elicit handling

[ihrpr]

sequenceDiagram
    participant User
    participant CLI as Chat CLI
    participant PA as Personal Assistant
    participant PRA as Project Analyst
    
    User->>CLI: "Help prioritize project issues"
    CLI->>PA: call_tool("chat", {message})
    
    PA-->>CLI: ElicitRequest(message, schema)
    Note right of PA: Need project name from user
    
    CLI-->>User: Display prompt & collect input
    User-->>CLI: Provide project name
    
    CLI-->>PA: ElicitResult(response)
    
    PA->>PRA: call_tool("analyze_issues", {project})
    
    PRA-->>PA: ElicitRequest(message, schema)
    Note right of PRA: Ask if okay to proceed
    
    Note over PA: Handle internally: auto-confirm
    
    PA-->>PRA: ElicitResult({proceed: true})
    
    PRA-->>PA: ToolResult(content)
    PA-->>CLI: ToolResult(content)
    CLI-->>User: Display result

Loading

[bhosmer-ant]

This is a really nice demonstration of

• how elicitation enables the core agentic interaction (worker querying supervisor for more info as task unfolds)
• the "agent = server + client" formulation

BTW one thing that was mentioned in modelcontextprotocol/modelcontextprotocol#382 that might be useful to add is an example of handling an elicitation from a subagent that needs to be forwarded up to the user - I suspect this pattern will be pretty common, and it fits within the intent of this example (rather than starting down the slippery slope of a more general "here's how to do agents" demo, which ofc this is very useful as a framing device for).