Too many MCP tools can make Claude Code harder to control
Running Claude Code in the terminal with about five connected is starting to create friction. Claude can see every connected tool in each session, so it may choose the wrong one or spend tokens checking a tool that was not needed.
The setup includes a MCP for the database, a MCP, and a few API wrappers. It works, but it feels assembled piece by piece without a clear plan.
The practical questions are whether should stay on all the time, whether they should be enabled per project, and whether CLAUDE.md should include rules for tool use. Adding more may eventually make Claude less focused instead of more capable.
Key points
- Claude Code can become harder to steer when many are always connected.
- Claude may pick the wrong tool or spend tokens exploring tools that were not requested.
- The setup includes , access, and several API wrappers.
- One option is to enable only for the projects that need them.
- CLAUDE.md can be used to give Claude clear rules for tool use.