Give a research agent a documented dataset with MCP

A research agent can find an answer only if it can reach useful data. dAgentBase exposes its catalogue through a remote MCP endpoint. Search, listing details, samples and quality reports are available without a key; claiming a free listing and getting its download URLs requires an account key.

1. Connect the remote endpoint

Create an API key in settings, then add the following remote server to an MCP client that supports URL-based connections. Keep the key private; it is shown only when created.

{
  "mcpServers": {
    "dagentbase": {
      "url": "https://dagentbase.com/api/mcp",
      "headers": {
        "Authorization": "Bearer dm_live_YOUR_KEY"
      }
    }
  }
}

The endpoint works without installing a package. See the MCP guide for client-specific setup and the current tool list.

2. Start with a real research question

Try: “Find historical S&P 500 constituent changes. Show me the source, dates, schema, licence and a few sample rows before you claim anything.” The agent can use search_listings,get_listing and preview_listing to evaluate the index constituent membership dataset. Read the displayed coverage and sample rather than assuming the dataset is fit for every backtest.

If it fits, ask the agent to claim the free listing and call get_download_urls. A signed URL expires after one hour; request another when you need the file again. The API quickstart shows the same search, preview, claim and download path with curl.

3. Check the data before using it

Inspect the source and licence on the listing, compare the latest data date with your research window, and run your own point-in-time checks. The A–F quality grade measures structural properties such as completeness and consistency. It cannot prove that every row is factually correct or free of look-ahead bias.

The current catalogue is free. A paid listing, when available, hands payment to a human Stripe checkout; agents cannot authorize paid purchases on their own. The remote endpoint is the supported connection path until the npm and Python packages are published.