Skip to content

adityachauhan0/CraftMCP

Repository files navigation

CraftMCP pixel mascot

CraftMCP

Open-source Windows design tooling for humans and AI agents.

Windows .NET 8 Avalonia Status

CraftMCP is a lightweight local design workspace where AI agents can create and update designs, and human users can inspect, edit, and export those same designs without leaving the workflow.

The project is intentionally focused. It is not trying to become a full professional design suite. It is trying to become a practical open-source tool people can pair with their own AI models for visual design work.

Why CraftMCP

Most design tools are optimized for direct human interaction, cloud collaboration, or very broad professional feature sets. That makes them awkward foundations for agent-driven workflows.

CraftMCP is built around a different set of constraints:

  • one canonical scene model,
  • the same model for human edits and agent edits,
  • deterministic exports,
  • local-first file handling,
  • and clear boundaries between UI, domain, rendering, persistence, and agent integration.

Intended Use Cases

  • Social graphics from prompts
  • UI mockups for desktop and mobile apps
  • Slide-style layouts and presentation visuals
  • Reopening an existing design and refining it with an AI model
  • Exporting structured design data so another agent can continue the work

MVP Principles

Principle Meaning
Human + AI collaboration AI can generate and modify designs, but humans stay in control and can edit directly.
Local-first Core editing and export should not depend on cloud services.
Deterministic output JSON export is a product contract, not a debug dump.
Lightweight scope Useful open-source tooling matters more than feature breadth.
Shared source of truth DocumentState remains the canonical scene model across editing, persistence, and export.

Current Progress

The project has moved well beyond the concept stage.

Verified foundation

  • .NET 8 solution scaffolded with explicit project boundaries
  • Avalonia desktop shell bootstrapped
  • Stable ID model added for documents, nodes, and assets
  • Core value objects and scene graph contracts implemented
  • DocumentState, CanvasModel, and hierarchy validation added
  • Initial test harness and domain tests landed

Verified product core

  • v1 deterministic JSON export contract implemented
  • DocumentJsonExporter added as the export boundary
  • Fixture-based export tests added for social, UI, and slide documents
  • Byte-identical repeat export behavior verified
  • Transport-neutral command contracts added for the command layer
  • DesignCommand, CommandBatch, CommandResult, and HistoryEntry implemented
  • Initial typed command families defined for create, update, delete, reorder, grouping, canvas, asset import, duplication, visibility, and lock state

Verified editor shell

  • Preset-based new document flow implemented
  • Open, save, save as, PNG export, and JSON export wired into the shell
  • Layer panel, property panel, canvas hit testing, zoom, pan, and direct manipulation added
  • Human edits route through the shared command engine and history stack

Latest implementation direction

  • The most recent implementation notes describe phase 7 work on the internal agent proposal loop
  • That work adds scene-context building, planner contracts, proposal review, approval, rejection, and activity-log provenance
  • The current local workspace is in active integration and the latest manual UI review note was blocked by a build failure in the dirty working tree

Last clean verification recorded in the knowledge base

  • dotnet test tests/CraftMCP.Tests/CraftMCP.Tests.csproj -> 66 passed, 0 failed
  • dotnet build CraftMCP.sln -> 0 warnings, 0 errors

Current Architecture Baseline

Area Current direction
App shell Avalonia + .NET 8
Source of truth Canonical DocumentState
Editing model Same scene model for human edits and agent edits
Persistence Native .craft package with bundled assets
JSON export Deterministic top-level artifact envelope
Agent model Internal prompt-driven command engine in MVP

Repository Layout

src/
  CraftMCP.App          Avalonia shell
  CraftMCP.Domain       IDs, value objects, scene graph, validation, exports, commands
  CraftMCP.Rendering    Rendering and export-facing composition
  CraftMCP.Persistence  .craft packaging and document hydration
  CraftMCP.Agent        Agent integration boundary
tests/
  CraftMCP.Tests        Domain, export, and command contract coverage
docs/
  engineering-conventions.md

What Exists Today

Today the repo already contains:

  • a working desktop shell,
  • foundational scene graph and document models,
  • deterministic JSON export infrastructure,
  • command-layer contracts for future execution and history behavior,
  • command-backed human editing flows in the workspace shell,
  • test fixtures and automated coverage for the core domain.

What is still actively being integrated is the review-first internal agent workflow on top of that editor foundation. The knowledge base shows that work as underway, but the latest local UI review was blocked until the workspace returns to a clean, buildable state.

Near-Term Roadmap

The next major implementation work is centered on command execution:

  1. command validation pipeline,
  2. transactional command execution,
  3. inverse generation and history behavior,
  4. undo and redo flows,
  5. finishing and stabilizing the review-first internal agent workflow,
  6. continued rendering and persistence integration around the stable scene model.

What CraftMCP Is Not

CraftMCP is not trying to be:

  • a Figma replacement,
  • a full vector illustration suite,
  • a cloud-first collaboration platform,
  • or an everything-in-one design product.

The scope stays intentionally narrow so the core human-plus-agent workflow can be strong.

Contributing

Contributions are welcome, especially in areas such as:

  • scene graph and document model design,
  • deterministic JSON serialization,
  • command schema and execution,
  • rendering/export consistency,
  • Windows-first UX,
  • local asset and package handling.

See docs/engineering-conventions.md for current repository conventions.

Vision

CraftMCP should become a practical open-source design workspace for people who want to work with their own AI models locally, generate a strong first pass quickly, and still retain full human control over the final result.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors