Search and preview marketplace datasets from ADK through a public MCP endpoint

[shagarwal]

I maintain DataBazaar, a marketplace where people and agents can list and retrieve datasets. Here is a small ADK example for inspecting a dataset before deciding whether to retrieve it.

The public MCP endpoint exposes three tools: search_datasets, get_dataset, and preview_sample. It needs no DataBazaar account or API key. Full dataset retrieval and purchases use the separate authenticated flow.

This standalone check discovers the ADK tools, searches for retail data, and asks for a sample of the first result. It makes no model call or purchase:

python -m pip install 'google-adk[mcp]==2.11.0' 'mcp==2.3.0'
python adk_demo.py

import asyncio
import json
import os

# This public endpoint does not use a Google client certificate.
os.environ['GOOGLE_API_USE_CLIENT_CERTIFICATE'] = 'false'
from google.adk.tools.mcp_tool import McpToolset, StreamableHTTPConnectionParams

def payload(result):
    if result.get('isError'):
        raise RuntimeError(result)
    return json.loads(next(c['text'] for c in result['content'] if c['type'] == 'text'))

async def main():
    toolset = McpToolset(
        connection_params=StreamableHTTPConnectionParams(
            url='https://api.databazaar.io/mcp/public',
            headers={'User-Agent': os.getenv('DATABAZAAR_DEMO_USER_AGENT', 'databazaar-adk-example/1.0')},
            timeout=30,
        ),
        tool_filter=['search_datasets', 'get_dataset', 'preview_sample'],
    )
    try:
        tools = {t.name: t for t in await toolset.get_tools()}
        found = payload(await tools['search_datasets'].run_async(
            args={'query': 'earbuds', 'category': 'retail', 'limit': 3},
            tool_context=None,
        ))
        rows = found['data']
        if not rows:
            print('No matching datasets; try a different query.')
            return
        dataset = rows[0]
        sample = payload(await tools['preview_sample'].run_async(
            args={'dataset_id': dataset['id']}, tool_context=None,
        ))
        print(json.dumps({'tools': sorted(tools), 'dataset_id': dataset['id'],
                          'title': dataset['title'], 'sample': sample}, indent=2))
    finally:
        await toolset.close()

asyncio.run(main())

Tested on Python 3.13.12 with Google ADK 2.11.0: all three tools were discovered; search returned a listing; the preview returned a CSV sample URL and schema. That verifies tool discovery and invocation, not an LLM/Runner workflow or a full dataset download. The sample URL expires. ADK emitted experimental-feature warnings for its auth/error-handling paths.

If you are building an agent that needs external data, try it against one real task. You can browse datasets, inspect a sample, then create an account to retrieve a matching dataset. If the data is missing, a clearly scoped bounty can state the fields, coverage, and budget you need. I would also welcome a reply describing a concrete dataset your agent cannot find.

MCP source · ADK MCP documentation