MCP
MCP 介面 · Connect AI Clients to Typed Research Tools
Use a compatible MCP client to request research operations, inspect tool inputs and results, and keep approval with the user.
什麼是 MCP？
Model Context Protocol (MCP) lets compatible AI clients request structured tool calls. StratCraft exposes typed research tools while preserving operation boundaries, evidence, and user control.
範例命令
Example research requests whose execution remains subject to tool contracts and user approval.
- 使用 RSI-14 策略對 AAPL 執行 2020 到 2024 年的回測
- 列出我按夏普比率排序的前 10 個策略
- Draft momentum candidates, then ask me before starting a backtest
- 顯示哪些策略的夏普比率高於 1.5 且最大回撤低於 15%
- Compare reviewed candidates with a transparent baseline and save the evidence
MCP 工具調用
# Claude can call StratCraft as a tool
run_backtest(
strategy="rsi_14",
symbol="AAPL",
period="2020-2024"
)
list_strategies(
sort_by="sharpe",
limit=10
)
generate_strategy(
description="momentum + mean reversion hybrid"
)
compose_portfolio(
method="frankwolfe",
top_n=50
)Compatible MCP Clients
Compatibility depends on the client and the enabled local tool contracts.
Claude（Anthropic）
GPT-4（OpenAI）
Gemini（Google）
透過 Ollama 的本地 LLM
設定
Configure a compatible MCP client for the local server.
Point the client at the local StratCraft MCP server and review the enabled tools and permissions before invoking research operations.