Search and preview dataset samples with MCPStreamableHTTPTool

[Shaurya Agarwal (shagarwal)]

I maintain DataBazaar, a dataset marketplace. Here is a small Agent Framework example for checking a dataset's metadata and sample before deciding whether to retrieve it.

It uses MCPStreamableHTTPTool with DataBazaar's public Streamable HTTP endpoint. Search, metadata, and sample previews need no DataBazaar API key. Full dataset downloads, including free listings, require an account; paid listings have a price. This example stops at discovery and preview.

Save the code below as discover.py, then run:

uv run --python 3.12 --with agent-framework-core==1.21.0 --with mcp==1.30.0 python discover.py

import asyncio
import json
import os

from agent_framework import MCPStreamableHTTPTool


def as_json(result):
    if isinstance(result, str):
        return json.loads(result)
    text = "\n".join(item.text for item in result if item.type == "text")
    return json.loads(text)


async def main():
    async with MCPStreamableHTTPTool(
        name="databazaar",
        url="https://api.databazaar.io/mcp/public",
        load_prompts=False,
        request_timeout=30,
        allowed_tools=["search_datasets", "get_dataset", "preview_sample"],
        static_headers={
            "User-Agent": os.getenv(
                "DATABAZAAR_EXAMPLE_USER_AGENT", "databazaar-agent-framework-example/1.0"
            )
        },
    ) as catalog:
        print("Tools:", [tool.name for tool in catalog.functions])
        result = as_json(await catalog.call_tool(
            "search_datasets", query="wireless earbuds", limit=1
        ))
        if not result.get("data"):
            print("No matching dataset; try another query.")
            return
        dataset = result["data"][0]
        preview = as_json(await catalog.call_tool(
            "preview_sample", dataset_id=dataset["id"]
        ))
        print(json.dumps({
            "title": dataset["title"],
            "listing": f"https://databazaar.io/datasets/{dataset['id']}",
            "preview": preview,
        }, indent=2))


if __name__ == "__main__":
    asyncio.run(main())

The example disables prompt loading because this endpoint exposes tools rather than prompt templates. allowed_tools lists the three discovery tools it uses. These are direct framework tool calls; no model provider or model call is needed for the example.

Verified on October 10 with Python 3.12.3, agent-framework-core==1.21.0, mcp==1.30.0, and HTTPX 0.28.1: the tool list contained all three tools, search found a wireless-earbuds retail dataset, and its returned sample URL served a CSV with 10 rows and columns including product_name, price, and retailer. The live catalog and sample URLs can change. A sample helps inspect the format; it does not establish the full dataset's accuracy or suitability.

If you have an agent that needs external data, try a query relevant to its task and inspect the sample. To retrieve a full dataset or publish data you own, the account and integration instructions describe the authenticated workflows. If nothing matches, you can also post a data request. What fields, geography, and time range does your agent actually need?

Public MCP source. This is an independent DataBazaar example, not an official Microsoft integration. The example and post were prepared with AI assistance; the direct framework calls and sample response were checked against the live endpoint.