MCP Connectors: Connect Claude, Cursor and VS Code to Simplio3D

Ask an AI assistant about your product configurator today and it answers from whatever you paste into the chat: a screenshot, a list of options, a price you typed out. MCP connectors remove that step. A connector gives the assistant a direct, permissioned line into your Simplio3D account, so it reads the real project (the option blocks, the conditional rules, the pricing formula, the part names inside the 3D model) and answers from that.

This guide explains what an MCP connector is, what a connected agent can and cannot do in Simplio3D, how risky changes stay under your control, and how to connect Claude, Cursor or VS Code.

What are MCP connectors?

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to outside tools and data. It has two sides. An MCP client is the AI tool you already use, such as Claude, Cursor or VS Code. An MCP server is the system the assistant connects to: it publishes a list of tools, each with a description and a schema, that the assistant can call when a question needs them.

A connector is the link between the two: the server’s address, plus a credential that says who is connecting and what they may touch. Claude calls it a custom connector; other clients call it an MCP server entry. For Simplio3D it is always the same two things:

• The endpoint: https://app.simplio3d.ai/mcp
• A credential: an OAuth sign-in that you approve on a Simplio3D consent screen, or a connection token you create under Dashboard → Integrations → AI Connections.

Once connected, the assistant sees Simplio3D’s tools alongside its own and decides when to call them from what you ask.

What a connected agent can do in Simplio3D

The Simplio3D MCP server exposes 115 tools: 51 that read, 61 that write under a controlled-write model, and 3 that run agent workflows. Between them they cover projects, 3D model inspection, materials, option blocks, conditional logic, pricing, forms, quote requests and store connections. In practice, a connected agent can:

• Read a real project: its models, option blocks and variants, materials, conditional rules, pricing blocks and formula, and form fields.
• Look inside the 3D model: part names and hierarchy, so it can check that every option still targets a mesh that exists.
• Audit a configurator: missing part targets, deleted materials, rules that can never fire and variants without a price.
• Explain logic and price a configuration: which rule hides an option, and what a given set of selections costs, calculated on the server.
• Build and edit: create projects, option blocks, variants, materials, conditional rules and form fields, and tidy up what is already there.
• Run agent workflows: start a multi-step job such as a project audit or a safe repair, answer its questions and follow its progress. Workflows pause for your approval and never publish.
• Summarise leads: read the quote requests you have received and what each customer configured.
• Answer product questions from the documentation: search the official Simplio3D docs and cite the page, instead of answering from memory.

Every tool, with the scope it needs and whether it reads or writes, is listed in the MCP tool reference.

What it cannot do

Some actions stay out of reach whatever scopes you grant. A connected agent cannot publish a project, create a share link or change embed settings. It cannot upload files or delete a project. It cannot connect or disconnect a Shopify or WooCommerce store, or read that store’s credentials. And it never receives database access, API keys, webhook secrets or share-view passwords: tools return curated summaries, not raw records.

Scopes: give each connection only what it needs

A connection can only see and call the tools its scopes cover. Start read-only and add one write scope at a time, as the job requires.

Several small connections are safer than one that can do everything. Give the assistant that reviews pricing its own connection with read scopes plus pricing write, and the one that builds projects another. Revoking one then never interrupts the other.

How changes stay under your control

Letting an AI edit a live configurator only works if a bad edit cannot reach your customers. Simplio3D handles every write in three steps: propose, confirm, snapshot.

• Low-risk edits go through without waiting for you, such as renaming a block or adding a variant.
• High-risk changes only ever produce a proposal: pricing, SKUs, deletions, bulk edits, shared materials and store product links. The proposal waits under Pending AI changes on the AI Connections page and lists every change it would make. You press Apply or dismiss it. No agent can approve its own proposal, and the rule cannot be switched off.
• Every applied change is snapshotted, so it can be undone, then re-validated and recorded in an audit log.

Proposals are short-lived. If the project has changed since the agent proposed something, the old proposal is refused rather than written over your newer edits; ask the agent again.

On top of that, a tool call succeeds only when four checks all pass: the account’s email address is verified, the workspace owner’s plan is active, the seat’s role allows the action, and the connection holds the scope. The narrowest one wins. A Viewer seat, for example, can never make or confirm a change, whatever scopes its connection has.

Connecting Claude without a token. In Claude, open Settings → Connectors → Add custom connector and paste https://app.simplio3d.ai/mcp. Claude finds Simplio3D’s sign-in on its own and opens a consent screen where you choose the workspace and the scopes. The connection then appears in the same AI Connections list, marked OAuth app, and you rotate or revoke it there like any other.

Prompts to try first

Name the project, and ask for evidence: “which rule hides it?” gets a sharper answer than “why is it hidden?”.

Limits worth knowing before you rely on it

• Agents read saved data. The editor autosaves a few seconds after each change, so an edit made moments ago may not be visible yet. Save, then ask again.
• Server-side prices and logic carry caveats. The price calculation and the rule evaluation cover the common cases, not every behaviour of the published configurator, such as per-unit Number Input pricing. Each result lists what it did not evaluate, and Preview and the published share view remain the final word.
• Rate limits. Each connection can make 300 requests every five minutes. A workspace accepts 30 applied AI changes every ten minutes, shared between every connected agent and the built-in assistant.
• Large answers are trimmed, and say so. A result over roughly 60 KB is summarised and marked as truncated, so the agent cannot mistake part of a list for all of it.
• What the agent reads goes to your AI provider. Quote requests include customer names and contact details, so grant the Quotes scope deliberately.
• Tokens expire after 90 days. Rotate one straight away if it has been pasted into a chat, a support ticket or a screen share. Rotating stops the old token at once.

One tool layer behind every AI surface

The MCP server is not a separate product bolted on. Simplio3D’s built-in AI assistant, the authenticated REST tool endpoint for your own automations, and external MCP clients all call the same tool registry, behind the same authentication, permissions and audit log. A capability behaves the same wherever you call it from, and a scope or role that blocks it in one place blocks it everywhere.

That is also how MCP differs from the REST API and SDK. The SDK is a typed client for your own code, authenticated with an API key, and that key is never accepted on the MCP endpoint. MCP is for AI agents, authenticated with a connection you scope per tool.