Model Context Protocol (MCP)¶
You can expose Dataiku agents and tools through the native MCP server, or build and host a custom MCP server using Code Studios. These tutorials cover both approaches.
Choosing an approach¶
Use the native MCP server to expose existing Dataiku agents and tools. Choose Code Studios when you need to write and host your own MCP server.
Native MCP server |
Custom MCP server in Code Studios |
|
|---|---|---|
Use case |
Expose Dataiku agents and tools through configuration. |
Implement your own MCP tools and server behavior in Python. |
Dataiku version |
|
|
Server implementation |
Provided by Dataiku |
Written by you using the MCP Python SDK |
Hosting |
Built into the Dataiku instance |
Code Studio published as a headless webapp |
Agent exposure |
Select the agent directly in project settings. |
Wrap the agent in an LLM Mesh Query tool, then call it from the server code. |
Configuration |
Instance policies and project selections |
Python code environment, server code, Code Studio template and webapp settings |
Endpoint |
|
|