Frequently asked questions

Pipeworx is an open MCP gateway that connects AI agents to live data from 389 sources — SEC EDGAR, FDA, FRED, the Census Bureau, USPTO, EPA, UN Comtrade, ClinicalTrials.gov, ATTOM, and many more. One gateway URL exposes 1,409 tools; the agent loads only the ones it needs per task.

MCP is an open protocol for connecting AI agents to external tools and data over JSON-RPC. It standardizes three primitives — Tools, Resources, and Prompts — so any MCP-compatible client (Claude Desktop, Claude Code, Cursor, VS Code, ChatGPT, Windsurf, etc.) can consume any MCP server. Pipeworx ships all three primitives plus seven gateway-native meta-tools.

Developers building AI agents that need live data, analysts using Claude/Cursor/ChatGPT for research, and teams shipping vertical agents (housing, fintech, pharma, govcon, agri, trade, ESG). Pipeworx is built for autonomous-agent traffic, not human-triggered workflow automation.

The MCP packs and tooling are open source on GitHub at github.com/pipeworx-io. The hosted gateway runs on Cloudflare Workers and is a managed product. You can run individual packs locally via npx @pipeworx/mcp-{slug}, or use the hosted gateway for the full catalog.

Open claude_desktop_config.json (in ~/Library/Application Support/Claude/ on macOS or %APPDATA%\Claude\ on Windows) and add: { "mcpServers": { "pipeworx": { "url": "https://gateway.pipeworx.io/mcp" } } }. Restart Claude Desktop. The tools indicator should show ~20 Pipeworx tools.

Run: claude mcp add pipeworx -- npx -y mcp-remote https://gateway.pipeworx.io/mcp. Or for a specific vertical: claude mcp add pipeworx-housing -- npx -y mcp-remote "https://gateway.pipeworx.io/mcp?vertical=housing".

In Cursor settings under MCP, add a new server with the URL https://gateway.pipeworx.io/mcp. In VS Code with the MCP extension, use the same URL in your settings. Windsurf accepts the same JSON config format as Claude Desktop. See /docs/getting-started for each client's exact steps.

Yes — ChatGPT supports MCP servers through custom GPT configurations and Project connectors. Point the MCP server URL at https://gateway.pipeworx.io/mcp. The same gateway URL works across every MCP-compatible client without modification.

No — most users connect via standard MCP client config (JSON or CLI). If you want to call Pipeworx from your own TypeScript/Node code rather than an MCP client, the @pipeworx/sdk package on npm provides a typed wrapper around the gateway.

389 packs spanning financial (SEC EDGAR, SEC XBRL, FDIC, CFPB, Alpha Vantage), economic (FRED, BLS, Eurostat, OECD, World Bank, IMF), regulatory (FDA, EPA, USPTO, Federal Register), scientific (ClinicalTrials.gov, RxNorm, USGS), trade (UN Comtrade, Census Trade, Treasury), real estate (ATTOM, Altos, HUD), government contracts (SAM.gov, USAspending, SBIR), plus weather, social, reference, and developer APIs. See /registry for the full list.

Every tool call hits the underlying source live, with caching governed by source-appropriate TTLs (e.g., SEC filings are cached briefly; FRED series for hours; static reference data for days). Every response includes _meta.cache.fresh_until so agents can decide whether to refetch. No scraped data, no AI-generated content — only primary-source institutional data.

No for public sources — FRED, SEC EDGAR, FDA, Census, BLS, EPA, ClinicalTrials, USPTO, UN Comtrade, and most others work without any keys. For premium sources like ATTOM, Altos, or Alpha Vantage, you can bring your own key via X-API-Key header (BYO tier) or use Pipeworx's metered access on paid plans.

SEC EDGAR appears within minutes of filing. FRED is daily for most series. NASS crop reports update monthly. FDA adverse events refresh quarterly (with the most recent quarter typically 2–3 months behind). Every response carries _meta.cache.fresh_until so the agent can reason about freshness before reusing data.

Call the pipeworx_feedback meta-tool with type="data_gap" and a message describing the source. The team reviews submissions daily and uses them to prioritize the roadmap. Common requests have led to packs being added within days.

Yes for moderate use. The free tier allows 50 calls per day anonymously (rate-limited by IP) or 2,000 per day with a free GitHub-linked account. Paid plans add unlimited usage with metered billing and access to premium verticals. All public-source data is included on every tier.

Daily window per identifier — IP for anonymous, IP + key hash for BYO, account ID for OAuth and paid. The limit resets at UTC midnight. Counters live in Upstash Redis with a 25-hour TTL. Bearer-token authentication takes priority over X-API-Key, which takes priority over anonymous fallback.

Stripe usage-based billing with metered events fired in ctx.waitUntil() after each paid-tier tool call (never blocking the response). The meter event name is api_call. Subscription lifecycle is handled by the /auth/webhook/stripe webhook. Sign up at /signup, upgrade from the /account page.

The gateway returns a structured error with _meta containing the limit, current count, retry-after time, and a hint to upgrade. The agent can read the metadata and back off, queue the call, or surface the limit to the user. No silent failures — every limit hit is visible.

The main gateway URL is https://gateway.pipeworx.io/mcp. Add ?vertical={slug} to scope to a domain bundle, or ?task={query} to surface the top-20 most relevant tools for a task. Per-pack endpoints are available at https://gateway.pipeworx.io/{pack-slug}/mcp.

The gateway exposes the full catalog and supports meta-tools (ask_pipeworx, discover_tools, entity_profile, etc.) that fan out across packs. Per-pack endpoints like /edgar/mcp expose only that pack's tools — useful when an agent should only have access to a single source.

Every response includes pipeworx:// resource URIs — for example, pipeworx://edgar/company/0000320193/filings or pipeworx://fred/series/UNRATE. Embed those URIs in your agent's output as durable citations; the resources/read MCP method fetches current content for any URI.

GitHub: github.com/pipeworx-io. The pipeworx_feedback meta-tool is the fastest channel to the team for bugs, feature requests, missing data, or praise — reviewed daily. Roadmap is signal-driven from feedback and telemetry.

Use the pipeworx_feedback meta-tool with type "bug", "feature", "data_gap", "praise", or "other" and a clear message. The team reviews submissions daily; submissions are signal-rated and shape the roadmap directly. For sensitive/security issues, email [email protected].