FocusAlpha Dashboards — stock screener, ETF screener, semiconductor news monitor, hedge fund 13F profile

Four working financial dashboards built on the FocusAlpha data API, published as open-source samples. Each is one self-contained HTML file. Open it in a browser and every control, sort and drawer works, because the data is baked into the page.

Live pages: app.focusalpha.ai/dashboards · GitHub Pages mirror: focusalpha.github.io/dashboards

Sample	What it shows	FocusAlpha tools behind it	Files
Global stock screener	Screen listed companies across 82 markets on guidance changes, filed events, ownership and fundamentals	screen_companies	sample · source · live source
ETF screener	Screen 5,607 US-listed ETFs by classification, expense ratio, NAV and premium, 20-day flows and measured holdings	screen_etfs	sample · source · live source
Semiconductor news monitor	News as labelled events (direction, impact, the names each story reaches) for 36 chip companies in the US, Taiwan, Korea, Japan and China	get_company_events, get_company_news, get_market_data	sample · source · live source
Hedge fund profile	What a manager says in its Form ADV brochure against what its 13F holds, with a replicating portfolio and peers	get_institution_profile, get_institutional_holdings, get_market_data, get_benchmark_prices	sample · source · live source
Onboarding tutorial	The guide page: the prompt to paste first, sample questions, what the data covers, and the three dashboards above with their Build-my-own prompts	—	page · source

The samples are frozen: data as of 2026-09-23

Every number in the index.html files is as of 2026-09-23 (the news monitor's events as of the evening of 2026-09-22, US Eastern; the ETF screener's daily rows through 2026-09-03, the newest valuation date in the dataset on the day it was frozen). Nothing in those files calls the API, so nothing in them will ever update. They exist to show what the data looks like and how a page can use it.

For current data, connect FocusAlpha in your AI (focusalpha.ai/start) and build the live version below.

Build your own live version

Each dashboard has two files:

• index.html — the sample, with a data snapshot baked in. Works anywhere, never updates.
• live.html — the same page with no data in it. Every number is read through the viewer's FocusAlpha connector (MCP) when the page opens. Small (60–115 KB), one file, no dependencies.

The Build my own button on every sample hands you a prompt for Claude. The prompt fetches live.html from this repository, publishes it as an artifact with the FocusAlpha connector declared, and then applies your conditions: your filters, your watchlist, your fund. You can do the same by hand:

1. Connect FocusAlpha in your AI: focusalpha.ai/start.
2. Fetch the live source, for example https://raw.githubusercontent.com/focusAlpha/dashboards/main/screener-dashboard/live.html.
3. Publish it as a claude.ai artifact with the connector declared, e.g. capabilities: { mcp: { servers: [{ server: "FocusAlpha", tools: ["screen_companies"] }] } }. Without the declaration window.claude.use('mcp') resolves null and the page cannot fetch.
4. Edit the one block that is yours: PRESETS[0] (stock and ETF screeners), SNAP.companies (monitor), PEERS (fund profile). Each live.html says so in a comment at the top.

Opened as a plain web page, without a connector, live.html shows only its empty state.

How the page reaches its data

Every page touches its runtime through one call, MCP.callTool(server, tool, input, opts) (the monitor also uses MCP.watchTool for the feeds). In index.html that call reads the snapshot baked into the file; in live.html it goes through the viewer's connector:

const mcp = await window.claude.use('mcp');            // null when no connector is available
const r   = await mcp.callTool('FocusAlpha', 'screen_companies', input, { cache: { staleTime: 120000 } });
const rows = r.payload.data;

Declare the connector under the name it carries in your tool list (focusalpha in Claude Code); call it at runtime by its display name (FocusAlpha).

Things the data will punish you for getting wrong

• Units differ by column. Margins, yields and growth are fractions (0.15 = 15%); price returns are percents (15 = 15%).
• A guidance raise is not good news. A raise means a number went up; good news means better for the company. A cost line raised is both a raise and bad news. The screener keeps them in separate columns on purpose.
• A 13F is not AUM. It is long US-listed equity only: no shorts, options, non-US listings or cash. A falling line can be a drawdown or a move into assets the form does not cover, and the filing cannot tell those apart.
• Multi-listed companies return one row per symbol per date from get_market_data. Filter by symbol before you compute a return, or you will average two exchanges.
• News direction and impact are a model's reading of an article, not reporting, not a measurement, and not a price forecast. Knock-on rows are inferences drawn from filed relationships.
• Filings are not in the news stream. 8-K, TDnet, DART and HKEX filings come through get_company_disclosures.

Licence

MIT. Take the pages, keep the attribution notice. The data inside them is public filing and market data served by FocusAlpha and is not investment advice.