MarketCanvas-Env

Minimalist 2D design canvas for RL agents and MCP.

A Gymnasium-compliant design environment exposing a design canvas (Text, Shape, Image) to agents via semantic JSON or pixel arrays.

Core Features

Dual Action Spaces: Semantic (API actions) and Low-level (Computer Use).
Hybrid Observations: Structured DOM-like JSON or raw RGB pixels.
Heuristic Reward Engine: Scores alignment, WCAG contrast, and constraint satisfaction.
MCP Native: Built-in server for LLM tool-calling interaction.

# Run the demo
python demo.py

# Start MCP Server
python server.py

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
engine		engine
env		env
rewards		rewards
specs		specs
tests		tests
.gitignore		.gitignore
.python-version		.python-version
README.md		README.md
WRITEUP.md		WRITEUP.md
demo.py		demo.py
main.py		main.py
pyproject.toml		pyproject.toml
server.py		server.py
uv.lock		uv.lock