Equibles

Docker Engine 20.10 or newer with Compose v2, 2 GB of free RAM (4 GB recommended, 8 GB if you enable the embedding stack), and about 5 GB of disk that grows as scrapers backfill history. Any modern x86-64 or ARM64 CPU works, and outbound HTTPS access is enough, so no inbound ports are needed. You do not install .NET or PostgreSQL locally, but the database must be ParadeDB rather than vanilla PostgreSQL because the migrations rely on the pgvector and pg_search extensions.

You must set SEC_CONTACT_EMAIL in the .env file, because SEC EDGAR requires a contact address in the User-Agent header. FRED economic indicators and FINRA short data each need a free API key from those providers; without them the matching scrapers are skipped cleanly while everything else keeps running. MCP_API_KEY and AUTH_USERNAME/AUTH_PASSWORD are optional and only add authentication to the MCP endpoint and web portal, which are open by default.

Point the client at the MCP server over HTTP at http://localhost:8081/mcp, or add an Authorization: Bearer header if you set MCP_API_KEY. Claude Desktop and ChatGPT Desktop take that URL in their MCP server configuration file, and for Claude Code you run claude mcp add equibles --transport http http://localhost:8081/mcp. Clients that launch servers as subprocesses can use the --stdio flag instead, which requires building the server first.

No. Stock prices come from Yahoo Finance as end-of-day OHLCV and refresh once every 24 hours, so the self-hosted build is for research and analysis rather than live trading. SEC filings are polled on a ten-second minimum interval but can take minutes to appear after publication, 13F holdings are quarterly with a 45-day filing lag, and FINRA and FRED data follow those agencies' daily or bi-monthly release schedules. Live quotes and options chains are cloud-only.

Yes. Set Worker__TickersToSync__0, Worker__TickersToSync__1, and so on to sync only a chosen list instead of every ticker, which keeps both disk use and backfill time down. Worker__MinSyncDate controls how far back the scrapers go: the default start is 2020-01-01, a later date makes the first sync much faster, and the earliest supported date is 2000-01-01. Both settings are read from the .env file on the next worker start.