MCP Query Playbook

Purpose

Provide reusable financial query playbooks so the router can execute consistent, auditable query chains. Tool selection and parameters must follow ~/Max/research-warehouse/docs/query-guide.md.

Each playbook defines:

question_pattern
primary_tool
confirmatory_tool
stop_condition
output_focus

Core Playbooks

1) 财报后异动归因

question_pattern: 财报后涨跌原因、业绩是否超预期、电话会关键信号
primary_tool: get_earnings_call
confirmatory_tool: query_options_flow, query_technicals
stop_condition: 出现 quality=error 或两次连续 partial
output_focus: 管理层表述、期权行为、价格结构是否一致

2) 产业链冲击传导

question_pattern: 上游涨价/断供如何影响核心公司
primary_tool: query_bottleneck
confirmatory_tool: query_supply_chain, query_company
stop_condition: 关键公司字段持续缺失
output_focus: 传导路径、受影响公司、活跃信号

3) 宏观冲击行业

question_pattern: 利率/通胀/风险偏好对行业或主题的影响
primary_tool: query_macro
confirmatory_tool: search_semantic
stop_condition: 宏观指标为空且语义检索为空
output_focus: 指标方向、宏观信号、历史可比叙事

4) 机构资金动向

question_pattern: 近期是否有机构资金偏好变化
primary_tool: query_etf_flow
confirmatory_tool: query_options_flow
stop_condition: flow 工具连续不可用
output_focus: 资金流强弱、持仓变化、异常合约

5) 个股风险排查

question_pattern: 某标的当前主要风险点是什么
primary_tool: query_company
confirmatory_tool: query_macro, query_technicals
stop_condition: 公司主数据缺失
output_focus: 基本面风险、宏观暴露、技术弱化信号

6) 研报观点验证

question_pattern: 某观点是否被数据支持
primary_tool: get_research_reports
confirmatory_tool: search_semantic, query_company
stop_condition: 无可用研报且检索为空
output_focus: 观点共识、反例证据、数据一致性

7) 技术面与基本面冲突诊断

question_pattern: 基本面看多但走势转弱是否合理
primary_tool: query_technicals
confirmatory_tool: query_company, query_options_flow
stop_condition: 技术数据为空
output_focus: 趋势结构、资金行为、基本面背离解释

8) 标的发现到深挖

question_pattern: 先找标的，再挑选深入研究对象
primary_tool: search_companies
confirmatory_tool: query_company, query_supply_chain
stop_condition: 搜索结果为空
output_focus: 候选池、基本特征、链条位置

Quality Guardrails

Any quality=error on primary tool: stop and repair input first.
Two consecutive partial: switch to "information insufficient" mode.
empty on primary tool: try one adjacent-tool fallback only.

Parameter Canonical Rules

query_supply_chain: use entity (not ticker), optional direction/max_depth
query_bottleneck: use domain in memory|photonics|packaging|power|gpu
query_options_flow: use ticker + optional days
query_etf_flow: optional ticker, optional weeks
query_technicals: use ticker + optional days

Output Focus Rules

State evidence first, inference second.
If evidence is cross-tool inconsistent, mark as unresolved.
Always include next actionable query when confidence is low.

Command-Level Examples (`rw`)

First configure once:

rw setup --server-url https://your-mcp-gateway.example.com
rw doctor

财报后异动归因

rw earnings --ticker NVDA
rw options --ticker NVDA --days 30
rw technicals --ticker NVDA --days 7

产业链冲击传导

rw bottleneck --domain packaging
rw chain --entity NVDA --direction both --max-depth 3
rw company --ticker NVDA

宏观冲击行业

rw macro --days 30
rw semantic --query "semiconductor demand under high rates" --limit 5

标的发现到深挖

rw search --sector photonics --tier chokepoint
rw company --ticker LITE
rw chain --entity LITE --direction upstream

mcp-query-playbook