ArchitectureA supervised JVM-class runtime — OLTP on seven engines, OLAP on three. AI-native, MCP-native, observable as plain SQL.Read the architecture
Está viendo la edición Perú. Está viendo la edición Colombia. You're viewing the Pakistan edition. Cambiar a la edición global →Cambiar a la edición global →Switch to the global edition →
AI and agentsProduct Release1 min read

The MCP server moves into the Studio runtime, with an SPI any module can extend

The protocol stack moved from the documentation module into the runtime, so any module contributes tools, resources and prompts through a service loader, and access is checked on every endpoint.

The MCP server began inside the documentation module, because the manual was the first thing worth exposing. By April 2026 it had outgrown its birthplace. This release moves it into the runtime and opens it to every module.

The move

  • Fifty-six files into the runtime The protocol types, the JSON-RPC dispatch, sessions and the endpoint moved into the web runtime; interfaces and transfer types moved to the AI library.
  • An SPI Sibling projects contribute tools, resources and prompts through the service loader. Results aggregate across providers, names prefixed by provider namespace. The nine documentation tools were the first to migrate.
  • Its own path The endpoint mounts at its own route, sessions clean up on close, and the handler registry can be read as a result set for inspection.

Access

  • Checked everywhere One access check across the POST, DELETE and event-stream endpoints, refusing otherwise. A dedicated user type for MCP clients, and a connector card in the user settings with a client selector.

Everything the blog has said about worlds, gates and audit since May stands on this move. Before it, there was one module's tools; after it, there was a platform surface.

On the platformAI & MCPPlatform

All posts