MCP Etf Holdings

mcp-etf-holdings

An MCP server for ETF data. Ask which funds hold a stock, compare fees, or pull top holdings. Data comes live from Yahoo Finance. No API key.

---

What you can ask

Search by company name or ticker symbol. "Which ETFs hold Nvidia?" works the same as "Which ETFs hold NVDA?". Ask about several funds at once and you get one comparison table.

---

Requirements

• Python 3.11+
• An MCP client: Claude Code, Claude Desktop, VS Code, Cursor, or Antigravity
• uv, which runs the server without a manual install

Install uv:

curl -LsSf https://astral.sh/uv/install.sh | sh

---

Setup

You don't need to install the server first. Every config below launches it on demand with uvx, in its own isolated environment.

To put it on your PATH instead, run pip install mcp-etf-holdings, then swap uvx mcp-etf-holdings for a plain mcp-etf-holdings in any config below.

Claude Code

claude mcp add etf-holdings -- uvx mcp-etf-holdings

Claude Desktop

1. Open your config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

1. Add the server:

   {

"mcpServers": { "etf-holdings": { "command": "uvx", "args": ["mcp-etf-holdings"] } } }

1. Quit Claude Desktop and reopen it. Closing the window is not enough.

VS Code

1. Create .vscode/mcp.json in your workspace. For a global config, run MCP: Open User Configuration instead.
2. Add the server. VS Code uses the key servers, not mcpServers:

   {

"servers": { "etf-holdings": { "type": "stdio", "command": "uvx", "args": ["mcp-etf-holdings"] } } }

1. Restart VS Code.

Cursor

Use the same JSON as Claude Desktop. Put it in .cursor/mcp.json for one project, or ~/.cursor/mcp.json for all of them. Restart Cursor.

Antigravity

Antigravity reads mcpServers from mcp_config.json. Put it in ~/.gemini/config/mcp_config.json for all projects, or /.agents/mcp_config.json for one:

{

"mcpServers": { "etf-holdings": { "command": "uvx", "args": ["mcp-etf-holdings"] } } }

In the IDE you can open it from the agent panel: MCP Servers → Manage MCP Servers → View raw config.

Or install it as a plugin from a checkout of this repo (the plugin is just the same config plus a manifest, in antigravity/etf-holdings):

agy plugin install /path/to/mcp-etf-holdings/antigravity/etf-holdings

For the IDE, copy that folder into .agents/plugins/ (one workspace) or ~/.gemini/config/plugins/ (all workspaces).

To use the opt-in HTTP transport instead, Antigravity names the URL field serverUrl: {"mcpServers": {"etf-holdings": {"serverUrl": "http://127.0.0.1:8000/mcp"}}}.

Verify it works

Ask your client each of these:

• "Which ETFs hold NVDA?" — calls find_etfs_holding_stock
• "Compare SPY, QQQ and VTI" — calls compare_etfs, returns one table
• "Which ETFs hold Nvidia?" — resolves the name to NVDA and says so

If you get real numbers back, you're done.

---

Example prompts

Copy any of these into a client that has the server connected.

Compare funds

Compare SPY, VOO, IVV and SPLG — they all track the S&P 500, so which is cheapest?

Compare QQQ, VGT, XLK and SMH on expense ratio and 5-year return.

Search by name or ticker

Which ETFs hold Nvidia?

What's the ticker for Berkshire Hathaway?

Find me the Vanguard total stock market fund and show its top holdings.

Find the cheapest exposure to a stock

What's the cheapest ETF to get exposure to Nvidia?

I want AMD exposure without buying the stock directly. What are my options, and what do

they cost?

Check a portfolio for overlap

I own VOO, QQQ and VGT. Am I doubling up?

Show me every holding that appears in more than one of SPY, SCHD and DGRO.

Discover funds by theme

Find semiconductor ETFs and compare the three biggest.

What dividend ETFs exist, and which has the highest yield?

Research one fund

Give me a deep dive on SCHD.

How concentrated is QQQ? What share of it is the top 5 positions?

Prompt templates

The server registers five prompt templates. Claude Desktop and Claude Code show them as slash-commands, so you don't have to write the prompt yourself.

---

Tool reference

Tools that return more than one row format the result as a markdown table.

etf_info(ticker)

Metadata for one ETF. For two or more funds, use compare_etfs.

• ticker (string) — ETF symbol, e.g. "SPY"

Returns name, category, AUM, expense ratio, dividend yield, NAV/price, and YTD / 3-yr / 5-yr returns.

compare_etfs(tickers)

Compares several ETFs in one table.

• tickers (array of strings) — e.g. ["SPY", "QQQ", "VTI", "SCHD"]. Case-insensitive and de-duplicated. Capped at 10 funds per call.

Returns Ticker, Name, Category, AUM, Expense, Yield, YTD, 3-Yr, 5-Yr. A ticker with no data still gets a row, plus a note naming it, so one typo doesn't discard the rest.

etf_holdings(ticker)

Top holdings of an ETF with portfolio weights.

• ticker (string) — ETF symbol, e.g. "QQQ"

Returns a ranked table of symbol, name, and weight %.

find_etfs_holding_stock(stock_ticker, limit, custom_etf_universe)

Reverse lookup. Finds which ETFs hold a given stock in their disclosed top positions.

• stock_ticker (string) — a ticker like "NVDA" or a company name like "Nvidia". Names are resolved to a ticker, and the output says which one it used.
• limit (int, default 20) — capped at 50
• custom_etf_universe (JSON string, optional) — search a specific list instead of the default universe, e.g. '["SPY","QQQ","XLK"]'

Returns matching ETFs sorted by the stock's weight, highest first.

Only top holdings are checked, roughly 10–15 positions per ETF. A stock held outside

those positions will not appear.

stock_exposure_summary(stock, limit)

The same reverse lookup, joined with each fund's cost and size. Use it for "what's the cheapest or largest way to hold this stock?".

• stock (string) — ticker or company name, e.g. "NVDA" or "Nvidia"
• limit (int, default 10) — capped at 25

Returns ETF, Name, Weight, Rank, Expense, AUM, sorted by weight in the stock.

lookup_symbol(query, limit, asset_type)

Resolves a company or fund name to its ticker symbol.

• query (string) — e.g. "Nvidia" or "Vanguard total stock market"
• limit (int, default 10) — capped at 25
• asset_type (string, default "any") — "any", "stock", or "etf"

Returns Symbol, Name, Type, Exchange.

search_etfs(query, limit)

Finds ETFs by name, theme, or category.

• query (string) — e.g. "semiconductor" or "dividend"
• limit (int, default 10) — capped at 25

Returns matching ETF tickers with names and exchanges. To resolve a *stock* name instead of finding funds, use lookup_symbol.

---

From source

For development, or to run without waiting on a release:

git clone https://github.com/vlearner/mcp-etf-holdings.git

cd mcp-etf-holdings uv sync

Without uv:

python3 -m venv .venv && .venv/bin/pip install -e ".[dev]"

The repo ships a .vscode/mcp.json pointed at your working tree, so VS Code picks up local changes with no extra setup. Point other clients at:

{

"command": "uv", "args": ["run", "--directory", "/absolute/path/to/mcp-etf-holdings", "mcp-etf-holdings"] }

Check that the server starts:

uv run mcp-etf-holdings

It will sit silent, waiting for JSON-RPC on stdin. That is correct for a stdio MCP server.

To poke at the protocol by hand:

npx @modelcontextprotocol/inspector uv run mcp-etf-holdings

HTTP transport (optional)

stdio is the default and what every editor config above uses. To serve over streamable HTTP instead:

uv run mcp-etf-holdings --transport http            # http://127.0.0.1:8000/mcp

uv run mcp-etf-holdings --transport http --port 9000

Point a client at it with { "type": "http", "url": "http://127.0.0.1:8000/mcp" } (VS Code, under servers). MCP_TRANSPORT, MCP_HOST and MCP_PORT set the same options.

It binds to loopback by default and has no authentication. Binding elsewhere is refused unless you pass --allow-insecure (or set MCP_ALLOW_INSECURE=1); put an authenticating proxy in front before exposing it beyond your machine.

Try it in Docker

A sample Dockerfile builds the server from this checkout:

docker build -t mcp-etf-holdings .

docker run --rm -e MCP_ALLOW_INSECURE=1 -p 127.0.0.1:8000:8000 mcp-etf-holdings

MCP_ALLOW_INSECURE=1 is required because the container binds all interfaces. Keep the 127.0.0.1: in the port mapping: it is what stops the container from being reachable by other machines on your network. The server logs a no-authentication warning inside the container; that is expected. This is for trying the feature, not a hosting recipe.

Tests

uv run pytest -q

The suite is fully offline. Both yf.Ticker and yf.Search are mocked, the latter by an autouse fixture, so no test can reach Yahoo by accident.

For coverage:

uv run pytest -q --cov=mcp_etf_holdings --cov-report=term-missing

Layout

src/mcp_etf_holdings/

server.py ← FastMCP entry point, 7 tools prompts.py ← prompt templates, exposed as client slash-commands fetcher.py ← async wrappers around yfinance + 24h TTL cache formatting.py ← markdown tables and shared number formatting top_etfs.py ← ~365 ETF tickers scanned for reverse lookups main.py ← enables python -m mcp_etf_holdings tests/ ← pytest suite, fully offline .vscode/mcp.json ← VS Code config pointing at the working tree pyproject.toml

Dependencies: mcp (protocol SDK), yfinance (data), pandas (holdings frames), httpx (transport).

---

Troubleshooting

---

Limits

Worth knowing before you trust a number:

• Only the top ~10–15 positions per ETF are published. Anything derived from holdings —