Skip to main contentSkip to footer
v1.0.3 · MIT

One plugin, every agent runtime.

Skills, three role agents, and one MCP connection install the same way in Claude Code, Claude desktop, Cowork, Gemini CLI, and Codex.

Key Facts

Version
1.0.3
License
MIT, public repo
Runtimes
Claude Code, Claude desktop app, Cowork, Gemini CLI, Codex, GitHub Copilot CLI
MCP
One remote connection, no local server
Account
Doctrine tools free; validators need an active Pro, Pro Plus, Teams, Enterprise, beta, or trial plan

Live now: Claude Code, Claude desktop app, Cowork, Gemini CLI, Codex, GitHub Copilot CLI. Not yet in Cursor's marketplace.

Same doctrine, one install per host.

Five skills, a remote MCP connection, and, in Claude Code, three role agents, all from one public repository.

Free doctrine tools, no account needed

Install in one step, per host

  1. Pick your host

    Select where you work. The install step adapts to your selection.

  2. Install the plugin

    Add the marketplace once, then install the plugin.

    Claude Code

    /plugin marketplace add aidesignblueprint/ai-design-blueprint
    /plugin install ai-design-blueprint@aidesignblueprint
  3. Try a free tool

    Paste this into Claude Code. It uses the free doctrine tools, so no sign-in is needed.

    First prompt

    Search the AI Design Blueprint principles for guidance on approval gates before an irreversible action.
  4. Unlock the validators

    The validators need a signed-in account with an active plan. Everything above keeps working without one.

Cursor does not list the plugin in its marketplace yet. Until it does, connect the MCP server by hand; the Cursor page has the steps.

The MCP server behind this plugin is listed on MCP Registry, Glama, Smithery and PyPI.

What's bundled

Skills

  • agentic-design-blueprint, doctrine for agentic UX, approvals, and human-in-the-loop workflows.
  • architect-validation-orchestration, the validate, consensus, and certify sequence for repeat reviews.
  • design-validation-orchestration, when to call design.validate and the accessibility floor.
  • spec-validation-orchestration, when to call spec.validate before anyone builds.
  • team-agents-orchestration, running PM, Designer, and Engineer together in one governed session.

Role agents (Claude Code only)

  • spec-pm writes or reviews the spec, graded with spec.validate.
  • spec-designer builds or reviews the UI surface, graded with design.validate.
  • spec-engineer implements or reviews the agentic code, graded with architect.validate.

Questions

Do I need an account?

No, not for the doctrine tools. Sign in only to call a validator, with an active Pro, Pro Plus, Teams, Enterprise, beta, or trial plan.

What leaves my machine?

The doctrine tools send your query to our server, which returns the doctrine and logs the call. They never read your files or repository. Feedback you choose to send is stored. The full flow, validators included, is on the trust and data handling page.

Can I use it in GitHub Copilot or Cursor?

GitHub Copilot CLI, yes: add this repository as a marketplace and install, as in the Copilot tab above. It is not in Copilot's curated plugin list. Cursor's marketplace does not list it yet; the Cursor page shows how to connect the MCP server by hand.

How do updates reach me?

Each host checks the marketplace on its own schedule. We publish once to the repository, every connected host picks it up independently.

What does the install marker send?

The MCP address carries one fixed marker naming only the plugin and its version, disclosed in the README. Each call also records the client name and software version your client declares, and a keyed hash of your IP address. Neither the raw address nor the raw user-agent string is stored.

How is this different from adding the MCP server by hand?

Same endpoint. The plugin adds the skills and, in Claude Code, the role agents in one install, instead of wiring each piece yourself.

Does it work in the ChatGPT app?

Yes, in Codex mode. Switch the ChatGPT app to Codex and the plugin's skills load with its MCP tools. A normal chat lists the tools but never calls them, so it answers from web search instead. Codex CLI works the same way.