A full-featured customization layer for AI assistants built on MCP. You get six element types that work together: personas that define behavior and enforce permission policies, skills that add discrete capabilities, templates with auto-derived variable substitution, agents that execute multi-step goals with state tracking, memories for persistent context, and ensembles that bundle everything into reusable configurations. Ships with 38 starter elements in a local portfolio directory. The dollhouse-expert-suite ensemble activates a guided setup persona. Create or edit elements in plain English, activate them to change what your AI can do and which operations require approval, then share back to the community collection. Tested heavily on Claude Code with varying support across Cursor, VS Code, Gemini CLI, and other MCP clients.
claude mcp add --transport stdio dollhousemcp-mcp-server uvx mcp-server