DollhouseMCP 2.0 | DollhouseMCP

# DollhouseMCP > Open-source AI customization through modular, composable elements. DollhouseMCP lets > you build personas, skills, templates, agents, memories, and ensembles as portable > YAML files — then snap them together like Lego blocks. Runs on the Model Context > Protocol (MCP) standard. AGPL-3.0 licensed. DollhouseMCP 2.0 is a local-first element system for AI customization. Elements live in ~/.dollhouse/portfolio/ on your machine, sync optionally to GitHub, and can be shared through the Dollhouse Collection community library. The server exposes a structured query layer (MCP-AQL) that reduces tool-definition token overhead by 85–96% compared to discrete tool definitions. The project is maintained by Dollhouse Research (dollhouseresearch.com) and published as @dollhousemcp/mcp-server on npm. ## Core Documentation - [Home](https://dollhousemcp.com/): Overview, quick-start install command, and feature summary - [Building Blocks](https://dollhousemcp.com/building-blocks.html): The six element types — personas, skills, templates, agents, memories, ensembles - [Agent Runtime](https://dollhousemcp.com/agent-runtime.html): How agents execute, the Gatekeeper permission system, and autonomy evaluation - [Dynamic Permissioning](https://dollhousemcp.com/dynamic-permissioning.html): Gatekeeper policies, allow/deny/confirm patterns, and per-operation approval flows - [MCP-AQL](https://dollhousemcp.com/mcp-aql.html): The structured query layer built on MCP — CRUDE routing, semantic endpoints, runtime discovery - [Portfolio Workflows](https://dollhousemcp.com/portfolio-workflows.html): Three-tier architecture — local portfolio, GitHub backup, community collection - [Platform Operations](https://dollhousemcp.com/platform-operations.html): Full platform surface — portfolio management, collection workflows, skill conversion, search, and execution lifecycle control - [Starter Elements](https://dollhousemcp.com/starter-elements.html): The 38 bundled elements in