Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@design-doc-generator-mcp create a design doc from my recent work"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
One command in Claude Code. A polished design doc — Markdown and HTML, both saved straight into your repo.
Real work gets done. The design doc doesn't — it gets skipped, or rushed into a note nobody reads.
Diagrams get left out because drawing them takes longer than the fix did.
Three months later, nobody remembers why it was built that way. Not even you.
Run /design-doc. Claude turns the session you just had into a structured, professional doc.
Two diagrams, generated and styled automatically. Nobody drags a single box.
The reasoning gets captured while it's fresh — not reconstructed from memory later.
Output | Where |
Markdown file |
|
Interactive HTML | Same folder, |
Both are always saved, every time — nothing to opt into.
Every doc includes:
Abstract, problem statement, architecture and flow diagrams, key decisions and trade-offs, risks — irrelevant sections are dropped, never padded
At least two Mermaid diagrams wherever the session supports them: a static architecture view and a runtime sequence view, sized and styled automatically
An auto-inserted table of contents
Two more doc shapes, detected automatically: a root-cause/bug-investigation report, and a comparison doc for "which approach" decisions
Related MCP server: adr-skills
npx design-doc-generator-mcp initRun once, inside the project you want to document — registers the MCP server and installs the /design-doc
skill. Restart Claude Code, do real work, then run:
/design-docFlag | What it does |
(none) | Drafts the default design doc for whatever was just discussed |
| Activates a Pro license key |
| Shows usage help |
Running it again on the same topic updates the same files — it never creates a duplicate.
This tool never reads your source files. It never calls an LLM itself. It only reads lightweight git metadata —
branch, git status, a git diff --stat summary, your last 15 commit messages. Claude drafts the document
inside your own session, on your own account.
Two things ever reach a network: the section template request when drafting (a mode string, plus your license
key if you've activated one), and your license key itself when you run --activate. Pro status is always
verified against that key server-side, never assumed from anything the client claims.
Free | Pro | |
Design docs | 5 / day | Unlimited |
Content depth | Full | Full |
HTML export | Included | Included |
Signup required | No | No |
Pro removes the daily cap. That's the only difference.
Elastic License 2.0 — free to install and use for drafting your own design docs. You may not resell it as a hosted service, or bypass its license-key checks.
This server cannot be deployed
Persistent memory for Claude Code and Cursor. Stop re-explaining your project every session.
Serves your design system and coding standards to coding agents, so they stop guessing.
Turn PRDs and product ideas into structured specs so coding agents build your intent, not theirs.
Persistent context for Claude. Your AI always knows your projects and next actions across sessions.
