MCP Server
An MCP server is a process that speaks the Model Context Protocol and exposes tools, resources, or prompts to Claude Code. Think of it as a plugin system: each MCP server adds capabilities that Claude Code doesn’t have out of the box.
How it works in claude-multi
Each instance stores its own MCP server configuration in settings.json inside its config directory. That means two instances can run completely different sets of MCP servers without interfering with each other.
claude-multi provides commands to manage MCP servers across instances:
claude-multi mcp listshows what’s configured whereclaude-multi mcp copycopies server configs between instancesclaude-multi mcp verifychecks that referenced binaries and paths still exist
Why it matters
Without per-instance MCP configs, every Claude Code instance shares the same ~/.claude/settings.json. You’d have to manually edit JSON files every time you wanted a different tool setup. claude-multi handles the plumbing so each instance stays independent.
Example
A database instance might run an MCP server that connects to PostgreSQL. A frontend instance might run one that serves design tokens. They don’t need to know about each other.
Related terms
- Instance Isolation
Each claude-multi instance runs in its own config directory with separate settings, history, MCP servers, and plugins. No shared state between instances.
- Plugin Sync
A mechanism that keeps plugins and skills consistent across claude-multi instances using symlinks, so you update once and every instance gets the change.
- Config Directory
The directory where a claude-multi instance stores its settings.json, conversation history, MCP server configs, plugins, and skills.