Workforce Data Explorer

One interface for 89 curated US workforce and labor market data sources — 30 with live API access spanning official statistics (BLS, FRED, Census, O*NET, DOL, SEC EDGAR) and real-time signals (Indeed Hiring Lab job postings, ADP payrolls, the Anthropic Economic Index).

Three ways to use it, easiest first:

What's inside

🖥️ The dashboard

workforce-data-explorer.streamlit.app — nothing to install.

• Catalog — search all 89 curated sources by topic

• Job Postings — real-time labor demand from Indeed, incl. AI's share of postings

• State Labor Markets — the four headline indicators for any state

• FRED / BLS / Census / O*NET / DOL / SEC pages — fetch, chart, and download any series as CSV

• AI Assistant — ask questions in plain English ("show me quits vs. openings since 2022"). The shared assistant allows a few questions per minute; paste your own free Groq key into the expander on that page for unlimited use.

🤖 The AI connector (MCP)

The best experience: plug live labor data into the AI you already use. The connector exposes 15 tools — FRED and BLS series, occupation wages, Census ACS, O*NET occupations, UI claims, SEC layoff/company filings, Indeed daily job postings, AI-postings share, state labor market snapshots, and AI-usage exposure by occupation — and your AI decides which to call.

Claude (claude.ai) — requires a paid plan:

1. Settings → Connectors → Add custom connector

2. Name: Workforce Data — URL: https://workforcemcp.beaconturn.com/mcp

3. In a new chat, ask something like "What's happening with layoffs at public companies this month?" or "Compare software developer and nurse wages."

ChatGPT — Settings → Connectors (developer mode) → add the same URL.

📍 The connector URL changed (July 2026). If you added it earlier as workforce-data-mcp.onrender.com/mcp, that address still responds for now but is deprecated and slower (it sleeps between uses). Point your connector at https://workforcemcp.beaconturn.com/mcp — the new home is always on, and the URL is permanent.

Claude Desktop app, running locally — the easy way is the extension bundle:

1. Clone the repo and create the venv (see Run it yourself).

2. Download Workforce-Data.mcpb from the latest release and double-click it.

3. Claude opens an install dialog — click Install, then point it at your cloned repo folder when asked. The connector appears in Settings → Extensions with an on/off toggle. Done — no config files.

⚠️ Edit this file only while Claude is fully quit (⌘Q). The app rewrites it from memory on exit, so edits made while it's running are silently lost.

{
  "mcpServers": {
    "workforce-data": {
      "command": "/path/to/workforce-data/.venv/bin/python",
      "args": ["/path/to/workforce-data/mcp_server.py"]
    }
  }
}

Claude Code — one command, available in every session:

claude mcp add --scope user workforce-data -- /path/to/workforce-data/.venv/bin/python /path/to/workforce-data/mcp_server.py

The hosted connector runs on an always-on server — no cold starts.

⚙️ Run it yourself

Requires Python 3.10+ (tested on 3.13). All dependencies are in requirements.txt.

git clone https://github.com/thelancehaun/workforce-data-explorer.git
cd workforce-data-explorer
python -m venv .venv && source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env   # then add your free API keys (see below)

streamlit run app.py            # the dashboard
python mcp_server.py            # the MCP server (stdio, for Claude Desktop)
python mcp_server.py --http     # the MCP server (HTTP, for remote hosting)

The app runs without any keys — SEC, O*NET, and parts of BLS/Census work keyless — but FRED (which powers most time series) and the AI Assistant chat need free ones.

API keys (all free, ~5 minutes total)

Deploy your own copies

• Dashboard → Streamlit Community Cloud (free): point it at your fork, paste your keys under app Settings → Secrets.

• MCP server → Render (free): the included render.yaml configures it — connect your fork as a Blueprint and add your keys as environment variables. Any Python host (Fly.io, Railway, a VPS) works too: python mcp_server.py --http.

Architecture

app.py                    Streamlit dashboard (Overview, AI Assistant, data pages)
ui_theme.py               Shared chart styling (validated palette, light/dark)
mcp_server.py             MCP server (stdio + streamable HTTP, TTL-cached tools)
workforce_data/
  catalog.py              Searchable metadata for 89 curated sources
  chat.py                 Tool-calling chat engine (Groq) + shared tool layer
  sources/
    fred.py bls.py census.py onet.py dol.py sec.py indeed.py aei.py

Fetched data is cached in memory in both apps (1 hour for time series, up to a day for searches and metadata), so repeat queries don't burn API quota.

Bugs, ideas, and working together

• Bug or wrong-looking data? Open an issue — templates make it quick.

• Want to suggest a data source? Same place — there's a template for that.

• Working together / press: the contact link is on the app's About page (kept off GitHub so scrapers don't get it).

License

MIT — see LICENSE.