--- name: active-etf-research description: Research supported Taiwan-listed and US-listed ETF holdings, portfolio changes, overlap, and stock ownership through the Active ETF member MCP. Use for factual ETF research and comparisons. --- # A repeatable research workflow The Skill helps your AI choose the right tools, check dates, and turn observations into a clear research note. MCP provides the data; the Skill provides the workflow. Covers supported Taiwan-listed and US-listed ETFs. Compare funds within one listing market; issuer data dates can differ. 13F portfolios are not included. MCP server URL: https://active-etf-mcp.inthewins.com/api/mcp Connections & usage: https://active-etf-mcp.inthewins.com/en/mcp/connect ## How the Skill works 1. Identify the ETF or stock and the observation period. 2. Use the smallest set of tools that answers the question. Prefer a single comparison or brief over repeated snapshots. 3. Check source dates, missing observations and differences between raw and scale-adjusted changes. 4. Present facts, uncertainty and source links in your language. Never turn missing data into zero or promise a return. ## Research tools - `search_market_entities`: Find an ETF (0 points). - `get_etf_snapshot`: Complete latest holdings (1 points). - `get_etf_changes`: Recent holding changes (1 points). - `compare_etfs`: Compare ETF holdings (2 points). - `get_stock_context`: Stock ownership across ETFs (2 points). - `build_research_brief`: Daily research brief (3 points). - `get_my_plan_usage`: Your remaining points (0 points). 20 points reset at 00:00 Asia/Taipei. All connected clients share one balance. Unused points do not roll over. Website browsing does not spend MCP points. Failed or empty queries do not consume points. Repeating the same query within 10 minutes returns the same cached snapshot without charging again. Maximum 20 calls per minute and 2 concurrent queries. Tool arguments: search_market_entities({query, market:"all"|"tw"|"us"}); get_etf_snapshot({code}); get_etf_changes({code, days:1..30, limit?:1..500}); compare_etfs({codes:[2..3 distinct codes from the same listing market]}); get_stock_context({symbol, market?:"tw"|"us"}); build_research_brief({codes:[1..3 distinct codes]}); get_my_plan_usage({}). US-listed ETF changes are bounded and may be truncated; weight changes are not necessarily trades. 13F portfolios are excluded. Connection setup: This Skill requires a separately configured OAuth MCP client. If tools are unavailable, guide the user to add the MCP URL in their client and complete browser authorization. For Codex CLI, use `codex mcp add active-etf --url https://active-etf-mcp.inthewins.com/api/mcp` and, when authentication is needed, `codex mcp login active-etf`. For Claude Code, save this file at ~/.claude/skills/active-etf-research/SKILL.md, add an absent server with claude mcp add --scope user --transport http active-etf https://active-etf-mcp.inthewins.com/api/mcp, then authenticate from /mcp or claude mcp login active-etf. Claude web uses Customize → Connectors for the remote MCP connection and does not load this local file automatically. Keep existing client settings. Use native OAuth discovery and authorization; never invent or edit authorization URLs, PKCE, state or callback parameters. If a link expires or fails validation, restart the native client connection to obtain a fresh link; do not keep refreshing the failed page. If browser login state expires, sign in again in the same browser. After connecting, call get_my_plan_usage({}) as a zero-point check. Let the client manage token storage and refresh; never place credentials in this Skill or chat. On invalid_token, let the client refresh first; if that fails, ask the user to reconnect through OAuth. On quota_exceeded, report resetsAt and stop retries. On request_in_progress or rate_limited, do not loop. Keep missing observations explicit. Repeated queries within the cache window return the prior snapshot. Never ask users to paste tokens in chat. ## Try asking - Compare 00981A and 00991A: shared holdings, different exposures, and source dates. - Summarize the last 7 days of holding changes for 00981A. Separate raw changes from scale-adjusted changes. - Which tracked ETFs hold 2330? Show their source dates and recent observed changes. - Compare US-listed ETFs DRAM and NASA: shared holdings, different exposures, and each issuer's data date. No trading, personalized buy/sell instructions or advertising. A data date is not a real-time quote. Company and fund names remain in their source language. Public-source research data. Not personalized investment advice.