DollhouseMCP Open-source AI customization through modular elements

Building Blocks

Six modular building blocks for AI customization.

DollhouseMCP treats AI customization as a system of reusable parts. Each element type does a different job, and they become more powerful when you combine them.

That is the core platform model from the server itself: when you activate a Dollhouse element, you are not just changing a prompt. You are changing what tools the AI can access, what commands it can run, and what operations require your approval.

Behavior

Personas

Shape behavior, tone, expertise, and priorities while also acting as security principals with permission policies.

Capabilities

Skills

Add discrete capabilities the AI can activate on demand, including code review, research, analysis, translation, and security work.

Structure

Templates

Standardize outputs with reusable structure, variable substitution, and repeatable formatting.

Execution

Agents

Execute multi-step goals with state tracking, autonomy evaluation, resilience policies, and a real execution lifecycle.

Context

Memories

Persist structured context across sessions so facts, project state, and preferences can accumulate instead of resetting.

Composed stacks

Ensembles

Bundle multiple elements into one activatable unit with activation strategy, conflict resolution, and coordinated permission policies.

These are not six boxes you choose between once. Use one or as many of these as you like, in any combination that fits the workflow you want to build, refine, activate, and recombine.

Browse Collection

What the files actually look like

The building blocks are stored as readable markdown or YAML in the local portfolio. The platform ships actual starter files that already model the structure, metadata, and execution patterns. 

---
name: "Debug Detective"
type: "persona"
description: "A systematic investigator specializing in troubleshooting and root cause analysis"
triggers: ["debug", "troubleshoot", "error", "investigate", "bug", "problem"]
version: "1.0.0"
author: "DollhouseMCP"
category: "development"
---

# Debug Detective

## Investigation Methodology
1. Evidence Collection
2. Hypothesis Formation
3. Testing Strategy
4. Data Analysis
5. Root Cause Identification
6. Solution Implementation
---
name: "Development Team"
type: "ensemble"
activation_strategy: "sequential"
conflict_resolution: "priority"
context_sharing: "selective"
resource_limits:
  max_active_elements: 10
  max_memory_mb: 512
  max_execution_time_ms: 30000
elements:
  - element_name: "technical-analyst"
    element_type: "persona"
    role: "primary"
    activation: "always"
  - element_name: "code-review"
    element_type: "skill"
    activation: "on-demand"
  - element_name: "task-manager"
    element_type: "agent"
    role: "coordinator"

Why this model matters

Reusable instead of disposable

You can refine a building block over time instead of rewriting the same instructions into every new session.

Combinable instead of monolithic

Behavior, capabilities, structure, and memory do not have to live inside one oversized prompt.

Readable as files

Elements stay understandable as markdown or YAML you can inspect, version, share, and recover.

Permission-aware

Active elements can shape the AI's permission surface, so customization affects execution control as well as tone.

Technical details that matter

Want to see what ships immediately? Browse the 38 bundled starter elements, explore the Collection for community-built elements, or read how dynamic permissioning changes with the elements you activate.