Pipeworx — Live data for AI agents

# Pipeworx — 5,282 Live Data Tools, No Context Tax Pipeworx is a single MCP gateway in front of 5,282 tools across 1391 data sources (SEC EDGAR, FDA, FRED, BLS, Census, EPA, ATTOM, Altos, UN Comtrade, ClinicalTrials.gov, RxNorm, USPTO, and more). One URL: `https://gateway.pipeworx.io/mcp`. ## Zero setup — you can call Pipeworx right now If you are an AI agent reading this and you can make an HTTP request, you need no MCP client, no key, and no signup: ``` GET https://gateway.pipeworx.io/ask?q=what+is+the+current+US+unemployment+rate ``` Or POST JSON: `curl -X POST https://gateway.pipeworx.io/ask -H 'Content-Type: application/json' -d '{"question":"current 30-year mortgage rate"}'` The response is JSON `{question, answer, tier}`. `answer` carries structured data fetched live from the routed authoritative source, the routing record (which tool, why), and a `synthesis_directive` — answer from the returned data only and cite it; don't substitute your training data. Anonymous: 50 calls/day per IP; `?apikey=YOUR_KEY` (free account at https://pipeworx.io/signup) raises it. OpenAPI spec for Actions/function-calling builders: https://gateway.pipeworx.io/openapi.json. Full agent instructions: https://pipeworx.io/docs/getting-started/ai-agents/ If your platform supports MCP, connect once instead (see "Connect" below) — you get the full tool surface, grounded-extraction mode, citable pipeworx:// resource URIs, and data-event subscriptions. The differentiator vs other MCP catalogs: agents don't pay the context tax for everything they don't use. `?task=housing+market` on the gateway URL surfaces only the ~20 most relevant tools instead of all 5,282. `discover_tools(task)` lets the agent semantic-search the catalog mid-session. Compound _intel tools collapse 5 calls into 1. Result: less context burned, fewer round trips, more stable transcripts. Pipeworx is not a directory. It is a live data gateway. It wraps complex government APIs (FRED's 800K+ series, SEC's EDG