Archi Automate

MCP for Revit: Revit MCP Server for AI Automation | Archi Automate

June 02, 2026

Short answer: MCP (Model Context Protocol) for Revit is an open, standardized way for an AI client such as Claude or ChatGPT to connect to a Revit MCP server, inspect the live model, and request operations against the Revit API. The Model Context Protocol defines how the AI asks for data and actions; a Revit MCP server like Archi Automate translates those requests into managed Revit transactions, with read-only, dry-run, and approval controls so nothing executes blindly.

If you have spent any time evaluating AI tooling for Autodesk® Revit®, you have probably hit the same wall: most integrations are brittle. A chatbot can describe a workflow, but it cannot reach into your open model and act on it. MCP for Revit changes that connection layer. This article explains what MCP is, what a Revit MCP server actually does, and where a controlled bridge fits into a professional BIM workflow.

What is MCP (Model Context Protocol)?

The Model Context Protocol is an open specification for connecting AI assistants to external tools and data. Rather than each application inventing its own private integration, MCP defines a common contract between an MCP client (running inside the AI assistant) and an MCP server (running next to the software you want to control).

Three concepts do most of the work:

Tools — actions the server exposes that the AI can invoke, each with a defined input and output. In a Revit context, a tool might inspect an element or compose an edit.
Resources — structured data the server can return to the AI for context, such as the contents of the active model.
Transport — the connection itself. MCP is open and inspectable, which means you can see what is being asked and answered, and swap clients without rewriting the server.

The practical benefit is separation of concerns. The AI client handles reasoning and language. The MCP server handles the real, typed operations against the host application. Neither has to know the internal details of the other beyond the protocol.

What is a Revit MCP server?

A Revit MCP server is the bridge that sits between AI clients and the Revit model. It receives requests over the Model Context Protocol, composes the matching Revit API operation, and returns results. The full chain has four clear hops:

AI client (Claude, ChatGPT, any MCP-compatible client) -> Model Context Protocol -> Archi Automate bridge (composes the Revit API operation) -> Revit transaction (executes against the live model)

Each hop has a distinct job. The AI client interprets what you asked for. The Model Context Protocol carries that as an open, inspectable request. The bridge composes the actual operation against the Revit API and checks it against the active policy. The model executes the approved work inside a managed transaction and returns what happened. Because the model itself is the source of truth, the AI is not guessing from a stale export — it is reading the live state of elements, parameters, and relationships before it acts.

This is fundamentally different from a fixed-command plugin. A traditional add-in ships a finite set of buttons. A Revit MCP server exposes the ability to compose operations across the Revit API surface on demand, which is why the same connection can answer a one-off audit question one minute and renumber rooms across fourteen floors the next.

What MCP can do in Revit

Once an AI client is connected through a Revit MCP server, the range of useful work is broad because it maps onto the Revit API rather than a preset menu. Common categories include:

Query elements — find every element matching a condition, by category, level, workset, or parameter value.
Read and modify parameters — inspect instance and type parameters, and write values to shared or project parameters where policy allows it.
Create views and sheets — generate elevations, apply view templates, and place views on sheets at a defined scale.
Export data — pull schedules, parameter sets, or audit results out to structured formats for coordination or reporting.
Run QA and code checks — cross-reference elements against naming conventions, completeness rules, or regulation packs and produce a punch list.
Execute multi-step tasks — chain inspection, decision, and edit steps that would otherwise need a custom script.

This is complementary to the tools you already trust. Dynamo, pyRevit, and C# macros remain excellent for known, repeatable workflows you have already invested in. A Revit MCP server covers the rest: the one-off questions, the audits, and the multi-step tasks that simply do not have a script yet. For a broader view of how these approaches sit side by side, see our overview of Revit automation.

How a Revit MCP server compares to a traditional add-in

The clearest way to understand the difference is to put the two models next to each other.

Aspect	Traditional Revit add-in	Revit MCP server
Interface	Fixed set of buttons and dialogs	Natural-language outcome, composed on demand
Coverage	Whatever was coded in advance	Across the Revit API surface, subject to policy
Best for	Known, repeatable workflows	One-off questions, audits, multi-step tasks
Client	Tied to the add-in UI	Any MCP-compatible AI client
Review	Depends on the add-in	Read-only, dry-run diff, and audit log by mode

Neither approach replaces the other. Scripted tools win where a workflow is already defined; an MCP server earns its place where no script exists yet.

Why MCP beats ad hoc AI integrations

Before MCP, connecting an AI model to Revit meant building a one-off integration: a custom API wrapper, a bespoke prompt format, and glue code tied to a single AI vendor. That approach is fragile and expensive to maintain. MCP improves on it in four concrete ways.

A standardized tool interface

The server publishes its tools in a defined shape. The AI does not need hand-written instructions for each operation; it discovers what is available and how to call it. When the protocol is the contract, the integration stops drifting every time a prompt changes.

Client compatibility

Because MCP is an open spec, the same Revit MCP server works with any MCP-compatible client. You can connect Claude today and a different assistant tomorrow without rebuilding the Revit side. As underlying language models improve, the bridge inherits the upgrade with no migration.

Separation of reasoning and execution

The AI reasons about intent; the server controls execution. This split is what makes the system reviewable. You can inspect exactly which operation the AI proposed before it touches the model, because composing and executing are distinct steps.

A cleaner permission model

A single, well-defined boundary is far easier to govern than scattered integrations. Permissions, scope limits, and audit logging live at the server, not smeared across prompts. That is the difference between an experiment and something you can run on a project.

What Archi Automate adds

Archi Automate is an MCP-powered AI automation layer for Revit. It is the bridge described above, built for professional use. Instead of choosing from a fixed list of commands, you describe the outcome in plain language. The connected AI agent inspects the live model, composes the required Revit API operation as dynamically generated C#, previews changes in dry-run mode, and executes approved tasks inside managed Revit transactions with automatic rollback on error.

The desktop side is a small console that runs next to Revit, with three screens: a Dashboard for connection, license, AI clients, and active Revit sessions; Guardrails for execution mode, safety limits, the activity log, and the raw policy file; and Connect your AI, which provides config snippets for Claude, Codex, and any MCP-compatible assistant. It supports Autodesk Revit 2025, 2026, and 2027 on Windows 10/11 x64, including multiple concurrent Revit sessions.

Two further capabilities matter for evaluation. First, multi-modal inputs: whatever the AI client can read becomes a Revit instruction — voice via compatible clients, site photos and sketches, PDF and DWG markups, and external data such as Excel exports, GIS, manufacturer catalogues, energy-model results, and building-code databases. Second, modular skills: you can package your studio's expertise — office naming conventions, view-template libraries, building-typology rules, and regulation packs such as TEK17, IBC, or NBC — and the AI combines them with the live Revit API. For step-by-step connection instructions, see our guide to Claude and Cursor Revit MCP setup.

Safety and review

An AI that can edit your model needs guardrails that hold up on a live project. A Revit MCP server should never be a blind automation button for irreversible changes, and Archi Automate is built around that principle with three execution modes.

Read-only — the AI can inspect any element, parameter, or relationship, but the bridge refuses all write operations and opens no transactions. This is the safe default for federated and review models.
Dry-run — the AI composes the operation, but execution stops at a per-element diff. You approve, edit, or discard each change, and the diff is exportable to JSONL.
Unrestricted — write operations run inside managed Revit transactions with automatic rollback on exception, plus enforced timeouts and API constraints.

Two safeguards are always on regardless of mode. Every composed C# snippet is screened against a configurable, pattern-based deny-list before it can execute. And every session writes to a per-session JSONL audit log that is replayable for incident review. Governance is set at the hub level, with per-role read and write modes, scope limits, deletion rules, and audit-log retention — for example, architects read-only, BIM leads dry-run, and a project director unrestricted.

The honest limitation is that this is a human-in-the-loop system by design. The AI composes operations and inspects results, but a person reviews and approves anything that changes the model in dry-run, and roles decide who is allowed to run unrestricted at all. That is the right trade-off for production work.

Example prompts

These are the kinds of natural-language commands a Revit MCP server can act on once connected:

"Cross-check every habitable room on Level 03 against TEK17 §12 daylight, ventilation and ceiling-height rules and produce a punch list."
"Take this consultant Excel of equipment loads, match each row to the corresponding electrical fixture and write the wattage onto the matching shared parameter."
"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."
"Renumber every room across 14 floors using the new tower-naming convention."
"Find all doors missing fire rating values and list them by level."
"Sketch five massing options for this 4,500m² site polygon."

To see how this connects to the wider picture of AI automation for Revit, the principle is consistent: describe the outcome, review what the AI proposes, and execute under policy.

FAQ

Is MCP the same as a Revit plugin?

No. A plugin ships a fixed set of commands compiled into Revit. MCP is an open protocol that lets an AI client request operations dynamically through a server. A Revit MCP server installs a small bridge next to Revit, but the value is the open, composable connection rather than a preset menu of buttons.

Does Revit have a built-in MCP server?

No. Autodesk Revit does not ship an MCP server out of the box. You connect one through a product such as Archi Automate, which runs a bridge next to Revit and exposes the Revit API to MCP-compatible AI clients under your own policy.

Can MCP modify a Revit model?

Yes, when permitted. In read-only mode the bridge refuses all writes. In dry-run mode it composes the change and stops at a reviewable diff. In unrestricted mode it executes inside a managed Revit transaction with automatic rollback on error. Roles and policy decide which mode applies.

Which AI clients can use MCP?

Any MCP-compatible client. Archi Automate works with Claude (Desktop and web), ChatGPT and GPT-based apps, and other MCP clients, with voice supported through compatible clients. Because MCP is an open spec, you can switch clients without changing the Revit side.

Is MCP safe for production projects?

It can be, with the right controls. Safe use relies on read-only defaults for review models, dry-run approval for edits, a pattern-based deny-list screening every composed operation, a replayable JSONL audit log, and per-role governance. It is a human-in-the-loop system, not blind automation of irreversible changes.

Get started with a Revit MCP server

If you are ready to connect an AI agent to your model under controls you actually trust, Explore Archi Automate for Revit and try it on a real workflow with a 14-day trial.

Related guides

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

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.