Quickstart

Go from data file to live MCP endpoint in under five minutes.

1. Sign up

Create a free account at /auth/signup. The free tier gives you one endpoint, 500 MB of storage, and 1,000 MCP calls per month — enough to evaluate end-to-end without a credit card. You can sign in with a password or with a one-time email link from /auth/magic-link.

2. Upload data or paste a URL

From the dashboard, click New endpoint and pick a source. The wizard accepts:

• Files — CSV, JSON, NDJSON, SQL dump, Excel, Markdown, TSV, TXT, HTML, plus PDF / DOCX / PPTX / images / audio via Workers AI.
• URL extraction — fetch a single page as Markdown, crawl an origin, or coerce a page into structured JSON.

Give the endpoint a name (optional — we'll auto-slug from the filename if you skip it) and submit.

3. Watch the live progress tracker

Provisioning runs asynchronously and typically completes in 30–60 seconds. The endpoint detail page shows a 7-step tracker that polls every 2 seconds:

1. queued — job dispatched to the ingest worker
2. fetching — pulling the file from R2 or rendering the URL
3. provisioning_db — creating a dedicated Neon Postgres project for this endpoint
4. loading_data — parsing and loading rows
5. introspecting — discovering tables, columns, primary keys, foreign keys
6. publishing — generating MCP tools and the public llms.txt
7. active — endpoint is live

4. Connect any MCP client

When the tracker hits active, the page reveals your endpoint URL, a freshly-minted bearer token (shown once — copy it now), and a config snippet for Claude Desktop.

{
  "mcpServers": {
    "products": {
      "url": "https://shipmcp.io/<your-slug>/mcp",
      "headers": { "Authorization": "Bearer smcp_..." }
    }
  }
}

Restart Claude Desktop. The agent calls initialize, reads the endpoint's instructions field (which points at the public llms.txt), then calls tools/list to see what's available — typically 5 to 8 tools per table.

What's next

• Connect Claude Desktop — and other MCP clients (Cursor, Continue, custom).
• Data ingestion — every supported format and what works best.
• Tool generation — what tools you get from your schema.