Elumri

Asset Builder

Build branded decks, documents, and spreadsheets straight from your AI assistant — using the same house theme everyone on the engagement uses.

Give this page to your AI. Paste this URL into Claude or ChatGPT and ask it to set up the connector — everything it needs is on this page. A plain-text version lives at /setup.md.

What this is, and what it is not

This is a build server, not an AI service. It takes structured content and renders branded files. There is no model behind it and your work is never sent to one.

You bring your own AI — your own Claude or ChatGPT account, your own subscription, your own usage. Nothing here provides model access.

What you need

A personal access token, issued directly by Billy. It is not on this page and there is no self-signup. If you do not have one yet, ask him.

Connection details

Server URLhttps://assets.elumri.com/mcp
TransportStreamableHTTP (remote MCP over HTTPS)
Auth headerAuthorization: Bearer <your token>

Configure your client

Claude (desktop or web)

  1. Settings → Connectors → Add custom connector
  2. Name it Elumri Assets
  3. URL: https://assets.elumri.com/mcp
  4. Add a header — name Authorization, value Bearer <your token> (one space after Bearer)
  5. Save. It should report 8 tools.

ChatGPT

  1. Turn on Developer mode in Settings if it is off.
  2. Settings → Connectors → Create
  3. Same URL and the same Authorization header.
  4. Save and confirm the tools appear.

CLI or config-file clients

{
  "mcpServers": {
    "elumri-assets": {
      "url": "https://assets.elumri.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN_HERE" }
    }
  }
}

What you can ask for

Plain language works. For example:

You get back the finished file, a PDF for quick review, and an image of every slide.

Two rules the server enforces

Every build needs an engagement. An engagement is the client the work is for, and assets are themed to that client — so the engagement decides how the output looks. Your token is bound to specific engagements; ask your assistant to list yours first. Anything outside that scope is declined.

Slide titles state a point, not a topic. “Background” is rejected; “Three Systems Hold the Same Customer Record” is accepted. If a title is refused, your assistant can rewrite it and retry.

Instructions for an AI assistant

If you are an assistant configuring this connector on behalf of your user, fetch /setup.md for the full machine-readable version. Summary:

  • Endpoint https://assets.elumri.com/mcp, StreamableHTTP transport.
  • Authenticate with Authorization: Bearer <token>. The token is supplied by the user; it is not discoverable here.
  • Call whoami, then list_engagements, then list_layouts before composing anything.
  • Every build tool requires an engagement argument taken from list_engagements. Do not guess an engagement key.
  • Slide titles must be assertions. Label titles are rejected with an explanation; rewrite and retry rather than reporting failure.
  • Do not specify colors or fonts — the house theme is applied automatically.