I want to be able to use Model Context Protocol to provide Cursor access to resources, similar to how Claude Desktop supports it.
For example by connecting the postgres MCP server, I could get the Cursor composer or chat to read from my DB autonomously.
See MCP Quickstart, the Announcement by Anthropic and the Sourcegraph blog post about how Cody AI has integrated it
I think the way Cody AI has integrated it would work best for cursor.
I second this…
third! would be so cool
looks like the exact way Cursor should function
pls dont fail behind Continue dev, they already implement MCP.
By the way - one way to do it now at home is to hack the language server and put mcp right there.
hint hint.
While I am not super up to speed on this one, it makes a lot of sense and should already be there in at least Composer.
Unless I don’t understand this correctly, file-level context allows you to feed specific files or even your entire codebase as context to the AI model. This helps the model understand the project’s structure, coding style, and dependencies, leading to more relevant and accurate code generation and suggestions.
I’d have to say some work still needs to be done here, but I have seen improvements.
While cursor retains your conversation history with the AI, providing context for ongoing interactions and allowing the model to “remember” previous requests and refine its responses based on your feedback, I think we can all agree that memory is a big issue if you are using this for anything meaningful, and don’t get me going on anything production level.
Is there some reason there isn’t an internal scheduler that reviews all of this regularly so that I (we) wouldn’t have to continue to provide context? I guess I am in the wrong thread now. Apologies, but what I am getting at is that I think cursor needs to spend more time fully completing what its fundamental offering is before taking on more. If that makes any sense…