/

/

How to install an MCP App — ChatGPT

How to install an MCP App — ChatGPT

The same connector URL, connected to ChatGPT. Add it once and ChatGPT can look up any part of your product, show it from any angle, and walk someone through a repair.

If you’ve already connected a scene to Claude, this is the same URL and the same two minutes — only the location of the settings changes. If this is your first connector, you don’t need the Claude guide first; everything you need is below.

A 3DScribe scene is your CAD model turned into something an assistant can answer questions from. Connect it to ChatGPT and it can look up any part by name, function or material, show it from any angle, pull real dimensions off the geometry, and walk a technician through a repair — with the model on screen.

A 3DScribe scene answering questions inside ChatGPT — the model on screen, not a description of it.

Before you start

  • A ChatGPT account that can add connectors. Availability differs by plan and changes over time; if you don’t see the screen described in Step 2, check your plan or ask your workspace admin.

  • A 3DScribe scene that has finished processing. A scene still analysing won’t serve a connector yet.

  • The scene’s connector URL. The same one you’d use anywhere else — a connector isn’t per-assistant.

Every connection authenticates. There is no anonymous access to a 3DScribe connector — whoever adds it will be asked to sign in.

Step 1 — Copy the connector URL

Open the scene in 3DScribe Studio and copy its MCP connector URL. It looks like https://mcp.3dscribe.com/s/your-scene-id.

Step 1 — the scene's connector URL in 3DScribe Studio. The same URL works in any MCP assistant.

Step 2 — Open Connectors in ChatGPT settings

In ChatGPT, open Settings and find the Connectors section. This is where ChatGPT manages the external tools it’s allowed to reach.

Step 2 — Settings → Connectors is where ChatGPT manages the external tools it can reach.

Step 3 — Add the connector

Choose the option to add a custom connector, paste the URL, and confirm. ChatGPT contacts the scene and reads back the tools it offers. Name it after the product or scene — you’ll want to tell them apart once you’ve connected more than one.

Step 3 — paste the connector URL into the custom connector dialog and confirm.

Step 4 — Sign in and approve

ChatGPT will send you through a sign-in step. Approve the connection and the connector becomes available in your conversations.

Step 4 — approving the connection. Every connection authenticates; there is no anonymous access.

Step 5 — Check that it works

Start a new conversation and ask something concrete: “Show me all the components in this scene.” You should get the component explorer — a real interface you can scroll and search — rather than a paragraph describing one.

Step 5 — the component explorer opens inside the conversation.

If that works, you’re connected.

What you can ask for

Thirty-four tools are available, but you never call them by name. You ask in plain language and the assistant picks:

  • “What’s this part?” Finds a component by name, function, material or description and opens a card with its photo, category and specs.

  • “Show me all the components.” A catalog you can scroll, search, and filter by assembly.

  • “Where does it sit in the machine?” Highlights the part in context and spins a 360° turntable of that assembly.

  • “How big is it?” A dimensioned diagram with measurements taken from the model geometry.

  • “What touches it?” Contacts, neighbours, and what has to come off first.

  • “Walk me through the repair.” A guided tour with the camera moving to each step.

  • “Show me one in brushed steel.” A render of the part in a described finish.

  • “What does the manual say about X?” Searches the documentation attached to the scene.

Answers arrive as interfaces you can scroll and click, not paragraphs describing them.

Ask concretely

The connector is strongest on specific references — a named part, an assembly, a procedure. “How does the pump come apart?” lands better than “what would cause a rattle?”, which asks the assistant to speculate past what the scene contains.

Who can see your scene

  • Public means any signed-in account can connect. It does not mean anonymous.

  • Private means the connection is checked against your organisation’s membership.

You can also switch off whole capability groups per organisation, so a customer-facing connector can offer browsing and tours without exposing render generation. Media is served through signed, expiring URLs.

Troubleshooting

The connector doesn’t appear

Start a new conversation — existing threads don’t always pick up a newly added connector. If it’s still missing, reopen Settings and check it shows as connected rather than pending.

You’re asked to sign in repeatedly

The approval didn’t complete. Remove the connector, add it again, and finish sign-in in the window ChatGPT opens rather than a separate tab.

“Not authorized” on a scene you can open in Studio

The scene is private and the account you’re signed into ChatGPT with isn’t a member of the owning organisation.

Images stop loading in a long conversation

Scene media uses signed URLs that expire. Ask the question again to get fresh links.

What’s next

  • Connect the same scene to Claude — the steps are identical apart from where Connectors lives.

  • Hand the connector to a customer or field technician and watch what they ask for.

  • Set up a creator connector, which adds eight more tools for editing and publishing scenes from the assistant.