Custom Agents - Datagate

Python Agent

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

server_params = StdioServerParameters(
    command="datagate-mcp",
    env={"DATAGATE_API_KEY": "dg_live_..."}
)

async with stdio_client(server_params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()

        # Discover datasets
        datasets = await session.call_tool("list_datasets", {})
        print(datasets)

        # Query
        result = await session.call_tool("query", {
            "dataset_ids": ["uuid-1"],
            "text": "machine learning transformers",
            "top_k": 5,
        })
        print(result)

TypeScript Agent

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";

const transport = new StdioClientTransport({
  command: "npx",
  args: ["-y", "@datagate/mcp"],
  env: { DATAGATE_API_KEY: "dg_live_..." },
});

const client = new Client({ name: "my-agent", version: "1.0.0" });
await client.connect(transport);

// Discover tools
const tools = await client.listTools();
console.log(tools);

// List datasets
const datasets = await client.callTool({
  name: "list_datasets",
  arguments: {},
});

// Query
const results = await client.callTool({
  name: "query",
  arguments: {
    dataset_ids: ["uuid-1"],
    text: "machine learning transformers",
    top_k: 5,
  },
});

When to Use MCP vs SDK Directly

If you’re building a custom agent and want full control over the response format (typed objects, error handling), consider using the SDK directly instead of MCP:

from datagate import DatagateClient

client = DatagateClient(api_key="dg_live_...")
datasets = await client.list_datasets()  # returns list[Dataset]
response = await client.query(text="...", dataset_ids=["..."])  # returns QueryResponse

result = await session.call_tool("list_datasets", {})  # returns formatted text

MCP is ideal when your agent framework handles tool discovery and invocation. The SDK is better when you want typed responses and programmatic control.