MIT
/open source
/C4 · DDD · governance
Architecture that lives here,
and holds whatever builds from it.
One model, built by you or by AI, that becomes the harness for whatever builds from it.
Modelled by either
Draw it in Studio, chat it with AI, or connect your agent over MCP.
Read as context
Agents read boundaries, ADRs and requirements before writing code.
Checked against
Fitness functions and requirements catch code that drifts from the model.
Also in the box
- Governance built in. Raw needs, EARS requirements, ADRs and fitness functions as linked model elements.
- AI that edits the model. Validated, undoable patches. Claude, OpenAI, Gemini or Ollama — your key, nothing proxied.
- MCP server. Claude Code, Claude Desktop or Codex read and edit a model folder with the same tools as Studio's assistant.
- Concept Hub. 120+ ready-made requirements, ADRs, patterns and blueprints.
- Smart Layout. Tries several layout engines and keeps the cleanest result.
- Design, view, present. Edit, explore read-only or turn views into slides.
-
Plain files, no lock-in. One
.radicalJSON file or a folder of Markdown — git-friendly. - Desktop, browser or VS Code. Same editor everywhere.
Start from the Hub
Skip the blank canvas — pick concepts and add them to Studio in one click.
Connect your agent
A local MCP server gives Claude Code, Claude Desktop, Codex or any MCP client the model as tools: read the summary and the metamodel rules, search, add and move elements, create views, sequences and slides, run Smart Layout. Every write is checked against the metamodel, and Studio shows it on the open canvas.
# Node.js ≥ 20, from the repository root
npm run build -w @radical/mcp
# point it at a Markdown model folder
claude mcp add radical-folder -- node \
/path/to/radical.tools/apps/mcp/dist/index.js \
--folder /path/to/my-model
Run it locally
Node.js ≥ 18 — or skip the install and use studio.radical.tools.
# clone, install, run
git clone https://github.com/radicaltools/radical.tools.git
cd radical.tools
npm install
npm run dev