How do you add MCP servers and plugins to an AI agent safely?
In Bandicot you install skills, plugins and MCP servers per project, and their tools run under the same approvals as everything else, on your Mac, with the permissions you grant. Before enabling one, check what it can read and write, where its credentials live and whether its provider sends data off the machine.
How it works in Bandicot
- 01Pick the kind
Skills carry reusable guidance, plugins add behaviour with an explicit lifecycle, and MCP servers expose tools through a protocol boundary.
- 02Review before you enable
Check what it can read and write, where its credentials live and whether its provider sends data off the machine.
- 03Let skills load on demand
Skills load when a task needs them and append to the end of the session, leaving the fixed prefix of every request alone.
Words you will see on screen
- Review plugin access
- MCP servers
- Requested capabilities
People also ask it as
- How to add an MCP server to an AI agent
- Are MCP servers safe to install
- AI agent plugins that ask before they run
- Load skills into an AI agent only when needed