How much access do you give Claude Code through MCP?
retained text
I’ve been adding a few MCP servers to Claude Code and started wondering how other people handle permissions. Do you usually give Claude access to every tool a server exposes, or disable the ones you don’t need? Reading issues or searching data feels fine, but deleting something, changing records, sending messages, etc. feels like a different level of access. Do you manage this per project or agent, or just approve things as they come up?
document fetched 2026-09-22 00:40 · canonical https://reddit.com/r/ClaudeCode/comments/1wmfcbr/how_much_access_do_you_give_claude_code_through