We’re excited to introduce something that fundamentally changes how you interact with your customer data: the Treasure Data MCP Server.
This new open-source tool connects Large Language Model (LLM) assistants like OpenAI, Claude, GitHub Copilot Chat, and Windsurf directly to your Treasure Data environment. With it, you can explore databases, analyze tables, and write SQL queries using nothing more than plain English.
And yes, it really works, and it’s already transforming how our teams and customers analyze data in real time.
So, what is an MCP Server?
First things first: the Model Context Protocol (MCP) isn’t your typical web server. You don’t host websites with it. Instead, think of MCP as a bridge, a secure, real-time communication channel between AI assistants and the tools and data they need access to.
The @treasuredata/mcp-server acts as a local process that gives your AI assistant a secure, structured way to connect with your Treasure Data instance. Once it’s running, your assistant gains superpowers: it can explore databases, describe tables, write and execute SQL, and even manage workflows like switching databases.
Why this changes the game
Data is only valuable if you can use it. The Treasure Data MCP Server dramatically lowers the barrier to entry for working with that data. Instead of waiting on a data team or learning SQL, anyone can now get fast, reliable answers in the tools they already use.
By combining the trusted scale of Treasure Data CDP with the conversational flexibility of LLMs, we’re giving teams a new way to work—one where insights are a conversation away.
Let’s talk use cases
The Treasure Data MCP Server is built for real-world workflows across departments.
1. Marketing analysts getting answers, fast
A marketing manager using Claude Desktop might ask:
“@treasuredata What were our top 5 most viewed products last week, grouped by region?”
In seconds, Claude translates that request into a Trino SQL query, runs it against Treasure Data, and delivers the answer. No SQL experience needed, no waiting on an analyst.
Example of Treasure Data MCP Server in action with Claude
2. Data engineers debugging tables on the fly
Need to check the schema of a new ingestion table? Just ask:
“@treasuredata Describe the events_staging table.”
You’ll get back the column names, data types, and nullability — without jumping into the UI or manually writing DESCRIBE statements.
3. Product managers exploring customer journeys
A PM in GitHub Copilot in Visual Studio Code can type:
“@treasuredata Using the user_journeys table, find paths that start with a product view and end with a purchase within 24 hours.”
The assistant builds the SQL, queries the data, and presents results — all from inside the code editor.
4. Security teams identifying suspicious activity
An analyst investigating suspicious behavior could ask:
“@treasuredata From the www_access table, show IPs that viewed more than 50 unique pages in under 10 minutes.”
Again, the assistant handles everything: it writes the SQL, uses TD_TIME_RANGE and TD_INTERVAL functions to filter time windows, and presents the results cleanly.
This type of analysis might take an hour of work normally—but with MCP, it becomes a minute-long conversation.
5. Deep diving into segments
The MCP Server can query parent segments, segments, and segment query definitions. A user might not know which parent segment contains meaningful data in dev. Claude Code and the MCP Server can help them find a couple of examples by massively issuing CDP API requests in parallel.
Easy setup, instant impact
To get started, you won't need any permanent installation. The most recent Treasure Data MCP Server version will be conveniently accessed through npx.
Then, just add a simple configuration to your AI assistant. For example, in Claude Desktop:
JSON:
{
"mcpServers": {
"treasuredata": {
"command": "npx",
"args": ["@treasuredata/mcp-server"],
"env": {
"TD_API_KEY": "your_td_api_key_here",
"TD_SITE": "us01",
"TD_ENABLE_UPDATES": "false",
"TD_DATABASE": "sample_datasets"
}
}
}
}
Once connected, you can say things like:
- “Switch to the customer_behavior database”
- “List all tables with ‘conversion’ in the name”
- “Show me a 5-row sample from purchase_events”
It’s that simple.
Tips to make the most of it
To really unlock value, try starting broad and iterating. Ask your AI to build a plan:
“Create a step-by-step plan to explore user engagement in the mobile_events database.”
It might respond by listing databases, identifying key tables, describing schemas, and suggesting queries. This guided approach gives you structure—while still letting you drill deeper based on what you find.
Also, remember that your AI can improve its own queries. After seeing initial results, you can say:
“Refine this to only include users in California.”
And it will adjust the SQL and re-run the query — all without you needing to touch code.
Get started now for free
Start by connecting your AI assistant like Claude, Copilot Chat, Windsurf, or another MCP-compatible tool. Then, let your imagination do the querying. No SQL? No problem. The AI figures it out.
Ask questions like:
- “What’s the top converting customer journey by country?”
- “Show me all tables related to customer transactions in the past 30 days.”
- “Generate a query to segment users with high lifetime value and low recent engagement.”
Whether you’re a data scientist, a marketer, or a product leader, the Treasure Data MCP Server makes working with your data as easy as having a conversation. Let your AI assistant talk to your data, and see what it says back.
Let your AI talk to your data, and discover what it’s been trying to tell you all along.
The server is available today in public preview on npm. It supports all major Treasure Data regions and is designed with enterprise security in mind.
Let’s build the future of data access together.
