Revit MCP with Claude, Cursor, and ChatGPT: AI Client Guide

Short answer: Revit MCP lets a range of AI clients — including Claude, ChatGPT, Cursor, and any MCP-compatible assistant — talk to Autodesk® Revit® through a single open transport. Archi Automate is the bridge that receives those requests, composes the Revit API operation, and runs it inside a managed transaction under your governance policy, so you connect the client of your choice rather than installing a separate plugin for each tool.

What this page covers

If you are evaluating AI automation for Revit, one of the first questions is practical: which assistant do I actually point at my model, and does each one need its own integration. This page answers that. It explains the client and bridge roles in a Revit MCP setup, lists the clients that are supported today with honest status labels, walks through example workflows per client, and sets out exactly what an AI client can and cannot do once it is connected.

The short version is that the AI client is interchangeable. The Model Context Protocol is an open specification, so the same Archi Automate bridge serves whichever compatible client you prefer, and your guardrail policy applies identically regardless of which one is connected. For a primer on the protocol itself, see our overview of MCP for Revit.

Client and bridge architecture

It helps to see the full path a natural-language request travels before anything changes in your model. There are four roles in the chain, and each one is replaceable or inspectable on its own.

AI client -> Model Context Protocol -> Archi Automate bridge -> Revit transaction

• AI client — the assistant you type or speak into (Claude, ChatGPT, Cursor, or another). It reads your prompt and any attached inputs and decides what needs to happen.
• Model Context Protocol — the open, inspectable, swappable transport that carries structured requests between the assistant and the bridge.
• Archi Automate bridge — the desktop component running next to Revit. It inspects the live model, composes the required Revit API operation as dynamically generated C#, checks it against your active guardrail policy, and screens it against a deny-list.
• Revit transaction — where approved work executes inside a managed transaction, with automatic rollback on error.

Because the protocol sits in the middle, swapping the assistant does not change the bridge, your policy file, or your skills. The intelligence at the top can improve over time while the controlled execution at the bottom stays exactly as you configured it. As underlying models improve, the same bridge inherits the upgrade with no migration on your side.

Supported AI clients for Revit MCP and status labels

We use plain status labels so you know what is documented as supported versus what simply works because the protocol is open. "Supported" means the product covers it directly. "Experimental" means the client speaks MCP and should work as any MCP-compatible client, but we are not asserting a hardened, dedicated integration. The point of building on an open spec is that any compliant client can connect to the same bridge.

On the Revit side, Archi Automate runs with Autodesk Revit 2025, 2026, and 2027 on Windows 10 and 11 x64, and supports multiple concurrent Revit sessions.

Example workflows by client

The prompt you write is the same regardless of client; what differs is how each assistant handles attachments and conversation. The bridge inspects the live model, composes the operation, and applies your policy in every case.

Claude Revit MCP

Claude Desktop is a common starting point because the Connect your AI screen ships a ready config snippet for it. It handles long multi-step instructions and attached files well, which suits model audits and documentation runs.

• "Cross-check every habitable room on Level 03 against TEK17 §12 daylight, ventilation and ceiling-height rules and produce a punch list."
• "For every unique exterior wall plane on the tower, create an elevation view, apply the standard view-template, and place it on a new sheet at 1:50."

ChatGPT Revit MCP

ChatGPT and GPT-based apps connect as MCP-compatible clients, a strong fit when you are pulling in external tabular data or working from images and sketches that the assistant can read.

• "Take this consultant Excel of equipment loads, match each row to the corresponding electrical fixture and write the wattage onto the matching shared parameter."
• "Find all doors missing fire rating values and list them by level."

Cursor Revit MCP and other MCP editors

Cursor is an MCP-capable editor, so it can target the same bridge as any MCP-compatible client. We label this Experimental: it should work through the open spec, but treat results as you would any unverified setup and lean on dry-run mode while you confirm behavior.

• "Renumber every room across 14 floors using the new tower-naming convention."

Custom and in-house clients

Because MCP is an open specification, a team can build its own client — a script, an internal tool, or a domain assistant — and connect it the same way. The bridge does not care which client originated the request; the same policy, deny-list, and audit logging apply.

• "Sketch five massing options for this 4,500m² site polygon."

Do I need a different setup for each client

No. You install and configure Archi Automate once. The Connect your AI screen gives you config snippets for Claude, Codex, and any MCP-compatible assistant, and the Dashboard shows which AI clients and Revit sessions are currently connected. Adding a second assistant is a matter of pointing it at the same bridge — your guardrail policy, your modular skills, and your audit retention settings stay in place. This is the core advantage of building on an open protocol rather than a per-vendor plugin: the work you do to standardize your studio's conventions and regulation packs is reused across every client.

Security and permissions: what a client can and cannot do

Connecting an assistant to a live model is reasonable to be cautious about. The controls do not live in the AI client — they live in the bridge, which is the only component that can touch Revit. Three things bound what any client can do.

First, the execution mode, set per role at the hub level:

• Read-only — the client can inspect any element, parameter, or relationship, but the bridge refuses all write operations and opens no transactions. Safe for federated and review models.
• Dry-run — the assistant composes the operation, but execution stops at a per-element diff you approve, edit, or discard. The diff is exportable to JSONL.
• Unrestricted — write operations run inside managed Revit transactions with automatic rollback on exception, under enforced timeout and API constraints.

A common policy is architects read-only, BIM leads dry-run, and a project director unrestricted, with scope limits and deletion rules attached.

Second, every composed C# snippet is screened against a configurable pattern-based deny-list before it can run, no matter which client requested it. Third, every session writes to a per-session JSONL audit log that is replayable for incident review. So the answer to "what can the AI client do" is: only what your role permits, only after deny-list screening, and always on the record.

Safety and review

None of this removes the BIM professional from the loop, and it should not. Treat dry-run as the default for any change you have not run before, read the diff the same way you would review a colleague's edit, and reserve unrestricted mode for work whose blast radius you understand. Irreversible or wide-scope changes deserve a human approving the diff, not a blind run. Dynamo, pyRevit, and C# macros remain the right tools for known, repeatable workflows; AI automation for Revit covers the one-off questions, audits, and multi-step tasks that do not have a script yet. The two approaches sit side by side.

FAQ

Can I use Claude with Revit through MCP?

Yes. Claude Desktop and Claude on the web connect as MCP clients, and the Connect your AI screen provides a ready config snippet. Claude Revit MCP is a supported setup.

Does ChatGPT work with Revit MCP?

Yes. ChatGPT and GPT-based apps connect as MCP-compatible clients to the same Archi Automate bridge, with your guardrail policy applied identically.

Can Cursor connect to Revit through MCP?

Cursor speaks MCP, so it can target the bridge as any MCP-compatible client. We label this Experimental rather than asserting a dedicated integration, so use dry-run while you confirm behavior.

Do I need a different setup for each AI client?

No. You configure Archi Automate once. Additional clients point at the same bridge, and your policy, skills, and audit settings are reused across all of them.

Is it safe to give an AI client access to my Revit model?

The client only does what your role-based execution mode allows. Read-only refuses all writes, dry-run requires you to approve each diff, every operation is deny-list screened, and every session is logged to a replayable JSONL audit file.

Connect your assistant to Revit

Ready to point your preferred AI client at a live model under your own policy? See Archi Automate MCP setup.

Related guides

Continue building out your AI-for-Revit workflow with Archi Automate for Revit and these related guides:

• MCP for Revit
• AI agent for Revit
• Control Revit with natural language
• Revit automation with AI

Archi Automate is an independent product by Archi Systems for use with Autodesk® Revit®. It is not affiliated with, endorsed by, sponsored by, or approved by Autodesk.