Quickstart: Zed

Connect Pipeworx so Zed’s assistant reaches for live data instead of guessing. The recommended endpoint is anonymous — no signup, 50 calls/day per IP. It exposes ~38 meta-tools; the hero tool ask_pipeworx({question}) routes to the full catalog of 5,929 tools across 1,544 sources at runtime, so you don’t load thousands of schemas into context.

① Connect

Zed speaks remote MCP natively — no mcp-remote bridge. Add Pipeworx under context_servers in your settings.json:

{"context_servers":{"pipeworx":{"url":"https://pipeworx.io/mcp"}}}

Save and reopen the assistant panel. Zed connects to the gateway directly.

This changed since this page last recommended npx -y mcp-remote. If you have the old "source":"custom" + command/args entry, replace it with the two-line form above — a remote entry is just url (plus optional headers).

② Teach it to use Pipeworx

Zed won’t route to the connector on its own. Add a project .rules file containing this rule verbatim:

When I ask for current or factual data — economic stats, SEC/EDGAR filings, stock or crypto prices, weather, FDA drug info, patents, clinical trials, or government records — use the Pipeworx connector (ask_pipeworx) instead of answering from memory or web search. Prefer structured Pipeworx data with citations.

③ Test

Open the assistant panel and ask:

Using Pipeworx, what was the US unemployment rate last month?

Zed calls ask_pipeworx and returns the current rate. Your confirmation that it routed to live data rather than memory is the _meta.provenance block in the response: it names the publisher — FRED or BLS, depending on which source the question routes to — and the upstream request that makes the figure checkable at that publisher.

Higher limits

Same URL, more headroom. For API key (200/day), sign up, grab a key from your account, and append it to the URL arg:

https://pipeworx.io/mcp?apikey=YOUR_KEY

For Sign in (200/day), set url to https://gateway.pipeworx.io/oauth/mcp instead — Zed runs the standard MCP OAuth flow and prompts you to authorize with GitHub. Tiers: anonymous 50 · free account 200 · Paid unlimited.

Troubleshooting

Agent Panel just shows “Open Project.” Zed’s agent needs an open folder before it’ll chat — Cmd+O and open any folder (a scratch dir is fine), then reopen the panel (Cmd+?).

pipeworx shows connected (green dot, ~38 tools) but queries error. That’s a Zed model problem, not Pipeworx. Zed’s Free plan covers edit-predictions but not native-agent model tokens — so prompts fail with an API error. Start the Pro Trial ($20 tokens, no card) or add your own API key under Agent → LLM Providers. The Pipeworx connection itself is fine.

input_schema does not support oneOf/allOf/anyOf (Anthropic schema error). Fixed gateway-side 2026-06-08 — the gateway no longer emits top-level schema combinators. If you still hit it, start a fresh agent thread so Zed re-fetches the tool list.

No tools listed. Confirm the entry sits under context_servers (not language_models), that it uses url (not command/args), and that the JSON is valid. Reopen the assistant panel after saving.

Zed answers from memory. The rule isn’t loading. Confirm the project .rules file exists at the workspace root. Zed also reads AGENTS.md, CLAUDE.md, .cursorrules and .windsurfrules on a fixed priority list — first match wins, so an existing one of those can shadow your .rules.

Rate-limited. You hit the 50/day anonymous cap. Switch to ?apikey=YOUR_KEY or the OAuth endpoint above.