餐具室MCP编排器
流程图LR U\[用户运行\\npython main.py\]-->|启动|A\[MCPApp\\n（dinner_party_orchestrator）\]
A -->|uses| O[Orchestrator\n(plan_type=full)]
O -->|calls| AR1[Agent: pantry_reader]
O -->|calls| AR2[Agent: menu_planner]
O -->|calls| AR3[Agent: shopping_list_writer]
O -->|LLM calls via\nAugmented LLM| LLM[(LLM\nOpenAI gpt-4o)]
subgraph MCP Servers
FS[filesystem MCP server\n(mcp-server-filesystem)]
end
AR1 -->|read pantry files| FS
AR2 -->|read pantry summary,\nwrite temp files if needed| FS
AR3 -->|write outputs| FS
subgraph Local Files
TASK[task.md]
PLAN[plan.md\n(generated)]
PANTRY[pantry/\npantry.txt]
OUTPUT[output/\n dinner_plan.md\n shopping_list.md]
end
A -->|reads| TASK
O -->|writes plan| PLAN
FS -->|reads| PANTRY
FS -->|writes| OUTPUT顺序图 参与者用户 参与者main.py 参与者协调人 参与者内裤阅读器 参与者菜单_计划 参与者shopping_list_writer 参与者FS作为文件系统MCP 参与者LLM
User->>main.py: python main.py
main.py->>main.py: load mcp_agent.config.yaml
main.py->>FS: start filesystem MCP (uvx mcp-server-filesystem)
main.py->>main.py: read task.md
main.py->>Orchestrator: create with 3 agents
Orchestrator->>LLM: plan steps for task
LLM-->>Orchestrator: plan (use pantry_reader, menu_planner, writer)
Orchestrator->>pantry_reader: execute subtask
pantry_reader->>FS: read pantry/pantry.txt
FS-->>pantry_reader: pantry contents
pantry_reader->>LLM: summarize pantry
LLM-->>pantry_reader: structured pantry summary
Orchestrator->>menu_planner: execute subtask
menu_planner->>FS: read pantry summary / task
menu_planner->>LLM: propose 3 menus + pros/cons
LLM-->>menu_planner: menus, missing ingredients
Orchestrator->>shopping_list_writer: execute subtask
shopping_list_writer->>LLM: draft dinner_plan + shopping list
LLM-->>shopping_list_writer: markdown content
shopping_list_writer->>FS: write output/dinner_plan.md
shopping_list_writer->>FS: write output/shopping_list.md
Orchestrator-->>main.py: final result text
main.py-->>User: done + files on disk