Introduction
Frame turns every model into analysis-ready datasets: an assets table with one row per element, a properties table with one row per element property and, when a saved view has its own report, a views table. The same Parquet files feed Frame’s Power BI and Excel templates.
You can hand those datasets to the AI you already use. There are two ways in:
- Remote MCP server. Add
https://www.bimframe.co/mcpto Claude, ChatGPT or any other client that speaks the Model Context Protocol. Sign in with your Frame account and the client can list your models, read dataset scopes and download the datasets with a column guide. - WebMCP site tools. Browser agents, such as the ChatGPT desktop app’s built-in browser or Chrome’s agent, can use the same tools directly from the Frame dashboard while you browse, once you turn on AI clients in your profile.
Both work on every plan, including Free, and neither uses AI credits. They are separate from the Frame agent, the terminal panel next to the 3D viewer that runs viewer commands on paid plans.
What a connected client can do
Every tool is read-only and only sees models that belong to the signed-in account.
| Tool | What it returns |
|---|---|
list_models |
The models in your library, their file types and whether a report exists |
list_model_scopes |
Whether a model’s datasets cover the whole model or a saved view with its own report |
get_model_datasets |
Download links and column schemas for the assets, properties and views Parquet files |
get_dataset_guide |
Column descriptions, the join key between assets and properties, and example queries |
The dashboard site tools add get_current_model, which identifies the model you
have open in the viewer, and create_viewer_embed, which asks you to confirm
before an embed link is created. A connected client cannot upload, change or
delete anything in Frame, and Frame never shares your Autodesk credentials, API
keys or account details through these tools.
Requirements
- A Frame account with at least one model that has a generated report. Report datasets follow the same plan rules as the Power BI integration: Free plans can use datasets from IFC models processed by the Frame engine; paid plans can use every model.
- For the remote MCP server: a client that supports remote MCP servers with OAuth sign-in. Claude (web, desktop and Claude Code), ChatGPT and most MCP-capable agent tools do.
- For WebMCP site tools: the latest ChatGPT desktop app with a plan that includes site tools (not available in Enterprise or Edu workspaces), using model GPT-5.6 Sol or GPT-5.6 Terra (GPT-5.6 Luna has WebMCP disabled), or another browser agent that supports WebMCP.
Option 1: connect over MCP
- In your AI client, add a new remote MCP server or custom connector and enter
https://www.bimframe.co/mcp. In Claude this is under connectors; in ChatGPT it is under connectors or apps; in Claude Code runclaude mcp add --transport http frame https://www.bimframe.co/mcp. - The client opens Frame’s sign-in page. Sign in with your Frame account and approve the connection. The consent screen lists exactly what the client will be able to read.
- Ask a question. The client chooses the tools it needs, for example:
- “Which models do I have in Frame?”
- “Download the assets dataset for Torre A and count the elements per category.”
- “What is the total wall area by level in the current version?”
- “Which properties are available on doors in this model?”
The client downloads the dataset files with its own tools and runs the analysis there. If the client’s sandbox cannot reach the internet, Frame also exposes each model’s datasets as MCP resources, including a single compressed SQLite file that Python can open without extra packages. Larger models take longer to download; the properties table can hold millions of rows.
Option 2: WebMCP site tools in the dashboard
Site tools use WebMCP, a proposed web standard that lets a website offer tools to an AI agent running in the browser.
- Sign in to Frame and open your profile from the dashboard sidebar.
- In the AI clients section, turn on Let AI clients read my model datasets. The setting takes effect immediately.
- Open the Frame dashboard inside the AI app’s browser. In the ChatGPT desktop app, open the built-in browser from the toolbar and go to the dashboard; the arrow in the address bar lists the available site tools.
- Ask about a model. “Describe the model I have open” uses the viewer context; the other questions above work the same way as over MCP.
Turn the setting off at any time to stop registering the tools.
What the AI app receives
- The list of your models with names, file types and whether reports exist.
- For a model you ask about, the download links and column descriptions of its datasets. Links point to Frame’s report artifacts and are not shown on the page.
- The dataset files themselves, retrieved by the AI app, which processes them under its own provider’s terms. Frame does not send data to AI providers on its own when you bring your own client, and none of this uses your AI credits.
Revoke access
- Remove the connector in your AI client to stop it from calling Frame.
- Turn off AI clients in your profile to stop registering site tools.
- Regenerate a model’s report to replace its dataset files, or delete the model. Old download links stop working within a day.
Troubleshooting
- The client cannot connect. Check the URL is exactly
https://www.bimframe.co/mcpand that you completed the sign-in in the browser window the client opened. - The address bar shows no tools. Confirm the AI clients setting is on, that you are using the desktop app’s built-in browser, and that the selected model is Sol or Terra.
- The client says no datasets exist. Generate a report for the model in Frame first, then ask again.
- The client says the plan does not include datasets. Report datasets follow the same plan rules as the Power BI integration.