Collaboration w/ Eclipse IDE MCP Tooling

[jwflicker]

There are some prototypes for bringing MCP and ACP support into the Eclipse IDE platform and perhaps some collaboration is possible with the Spring Tools team with the recent addition of an MCP server to https://spring.io/tools#eclipse

There is a potential proposal for adding extension points to declare MCP tools, resources and prompts using annotated java functions via: https://github.com/spring-ai-community/mcp-annotations, and perhaps for adding individual MCP server endpoints, and contributing a set of platform MCP tools such as working with files, markers, annotations, ...

https://github.com/eclipse-agents/eclipse-agents/

Feel free to join a matrix chat thread for discussing such potential with Eclipse IDE committers:

https://chat.eclipse.org/#/room/!DquClGHIDXMzJfsxUy:matrix.eclipse.org/$9bxRGFRNy89fm7WFolEFC4pqHfPL7hF3GlT4-0Zm-aU?via=matrix.eclipse.org&via=matrix.org&via=beeper.com

[jwflicker]

FYI: @martinlippert

[martinlippert]

Thanks a lot for reaching out, happy to integrate and connect the AI related features that we are building with your efforts here.

Looks like we need to explain things in more detail here. The Spring Tools for Eclipse include a language server that contains most of the tooling features that we provide for Spring projects, so we try to reduce the amount of code that we need to write specifically for one of the IDEs that we integrate with. From the AI feature perspective, this means:

1. the MCP server that we ship is embodied in the language server that we built. The MCP server runs in HTTP/SSE transport mode. It gets started as part of the language server (and shutdown the same way) and opens a port on localhost, so that clients can connect and use this MCP server.

2. we have shortcuts showing up as code lenses in source code that trigger chat integrations with the surrounding AI coding assistant. This connection is implemented on the client side via a command that is executed when the user clicks on a code lens. This command then basically sends a dynamically crafted prompt to the AI chat interface.

For 1), the MCP and ACP support that is coming to Eclipse would be the place where users would then go ahead and configure the Spring Tools MCP server, if they want to use it. We could think about ways to make this easier (by automatically configure the MCP server) via an API that the Spring Tools would invoke.

If we go down that path (the platform provides an API to configure the MCP server), it would be important to design this API in a way that we could configure and change the MCP server entry, e.g. if the user enables our MCP server in the preferences or changes the port. It would also be good if the API provides a way to check whether a certain MCP server is already configured, so that we don't configure the MCP server on every startup, but could change the configuration if a preference have changed (in the meantime).

For 2), it would be good if the chat client UI would provide an API for us to send and execute prompts to.

That is - at least - what comes to my mind initially here. WDYT?

[martinlippert]

As a side note: we need to keep in mind here that we can't easily add bundle dependencies, so things need to be optional. Our overall approach is to integrate with whatever users decided to use, not to prefer one specific AI assistant.