I want to set up a Cursor automation that uses the Vercel MCP to periodically review the Vercel logs, notify in Slack, and propose code fixes.

However, it appears like the MCP permissions are all or nothing. Essentially, I have to TRUST that the LLM isn't going to delete production (or do any one of the dozens of disastrous things you can do with the MCP).

When I use the Vercel MCP in Cursor, it's safe: I can disable the dangerous tools with fine-grained control.

What can you do to make it safe to use the MCP in a cloud automation context?