Not a high-priority request, but sometimes I come across a MCP server that I would like to use, but I won’t use it because it has tools within it that I’m not convinced that I would trust out of the box.
This isn’t necessarily a security concern, more of a concern that it would make unnecessary changes that are hard/impossible to recover from.
For instance: a Jira server MCP that creates tickets, updates descriptions, etc - I would probably disable these initially so that I could test and build trust that the Agent uses it appropriately and doesn’t create terrible tickets or make ticket updates in error.
Being able to disable specific tools would reduce a certain amount of friction by letting me use certain ‘read-only’ tools, whilst I test the other tools in a separate instance, targeting a test environment.
This is easy to solve because the mcp implemented in vscode 1.99 (the latest version) is better and more user-friendly than the one implemented by cursor. Cursor only needs to develop based on it, as the current version that cursor is based on is 1.96.
This one seems like it should be critically important for Cursor, especially with the proliferation of closed source MCPs. Other MCP clients can do this.
I believe this was implemented in v0.50. Once it rolls out you can use this to disable/enable each tool of all the MCPs you have installed. Beta testers (EAP) can use it once installed.
