Connect Model Context Protocol servers to models accessed through OpenDunes.
Not yet available. Native MCP server orchestration through the OpenDunes API is tracked to be built — today, run MCP servers yourself and drive the tools loop with live tool calling on the chat completions API (see below). This page describes the planned integration.
The Model Context Protocol (MCP) lets AI models call external tools — file systems, databases, APIs — through a standardized interface. Once the OpenDunes MCP layer is live, you'll be able to attach MCP servers to any model in the catalog and have usage billed in DA.
MCP sits between your client and the model. Your client connects to one or more MCP servers, which expose tools. When a model decides to call a tool, the request routes through the MCP layer:
You'll declare servers and grant the model permission to call them in the request:
mcp_servers parameter on POST /v1/chat/completions.You don't need to wait — you can run MCP servers locally and orchestrate them yourself using the OpenAI SDK. The model handles tool-calling; your code executes the tool and feeds the result back as a tool role message.
Tool calling is live today on POST /v1/chat/completions — pass tools and tool_choice exactly as you would with the OpenAI API. See Tool Calling for the full request shape.
Watch the OpenDunes changelog for the MCP integration announcement.
tools capability.