Memoire MCP: Design System Diagnostics Through the memi CLI

What is Memoire?

Memoire, shown in the README as Memi, is described as the design layer for agentic AI. It maps the UI already in your repository, surfaces accessibility and design system risks, and gives a deterministic check to rerun before merge.

It ships as the npm package @memi-design/cli, which installs the memi command. The MCP server is one command of that CLI, and the 2.8 beta starts it as locked stdio, without sockets or project writes.

Ways to run it

• Install the CLI globally with npm i -g @memi-design/cli, then run memi mcp start --no-figma.
• Run a one off audit with npx -y @memi-design/cli@latest diagnose . --json --no-write --fail-on none.
• Give an agent the audit skill with npx skills add memi-design/memi --skill audit-frontend-design.

The first audit needs no account, API key, Figma file, global install or daemon. Local source diagnosis uses deterministic rules, so it makes no model call and carries no model fee.

What the audit returns

The diagnose result carries normalized finding IDs, confidence, provenance and file:line evidence, so an agent can act on a specific finding instead of guessing. The same check can run on every pull request through a GitHub Action.

Who it suits

Best for

• Teams that run coding agents and want findings with file and line evidence.
• Pull request checks that fail only on newly introduced interface debt.
• Agents that load skills, since the README gives an audit-frontend-design skill installed with one npx skills add command.

Not for

• Broad site audits across performance, SEO and best practices, which suit web quality audit. Memoire diagnoses the repository you provide.
• Projects that want a hosted endpoint, which Design Systems MCP offers. The documented path is a local CLI process.
• Setups that need the four locked read tools on the stable release. The README places them in the 2.8 beta, and stable remains 2.7.9.

Connect Memoire MCP: Design System Diagnostics Through the memi CLI to your AI client

memi mcp start --no-figma

1. Install the CLI

   Run npm i -g @memi-design/cli so the memi command is on your path. For the four locked read tools, install the exact @memi-design/[email protected] package.

2. Try the diagnose check

   Run npx -y @memi-design/cli@latest diagnose . --json --no-write --fail-on none in a project. The README calls it a non-destructive audit.

3. Register the MCP server

   Add a stdio entry that runs memi mcp start --no-figma to your MCP client config, using the JSON above. The README lists this MCP row as working with any MCP client.

4. Or use the agent installer

   Run memi agent install claude-code --project . for Claude Code. The README lists the same command for codex, cursor and grok-build.

Client configuration

{
  "mcpServers": {
    "memoire": {
      "command": "memi",
      "args": ["mcp", "start", "--no-figma"]
    }
  }
}

Endpoints and ports

What you can do with it

• CLI firstEverything starts from the memi command, installed from the @memi-design/cli package.
• Local MCP servermemi mcp start --no-figma starts a server on your machine, and the README lists it as usable from any MCP client.
• Non destructive auditThe diagnose command with --no-write runs one audit that needs no account, API key, Figma file, global install or daemon.
• File anchored findingsResults carry normalized finding IDs, confidence, provenance and file:line evidence.
• Locked read toolsThe 2.8 beta makes four read tools available without grants, and locked stdio starts without sockets or project writes.
• Pull request gateA GitHub Action fails only on newly introduced interface debt and uploads a memi-design-health report artifact.
• MIT licensedThe repository carries an MIT license.

Tools it exposes

Frequently asked questions