Developers

Connect Claude, ChatGPT or your editor

The hosted server needs no key: the client signs you in to AIBIMTool, and what it builds is billed to your own credits. Part of Studio. Server URL: https://aibimtool.com/mcp

In claude.ai or Claude Desktop: Settings → Connectors → Add custom connector. Name it AIBIMTool and paste the URL:

https://aibimtool.com/mcp

Connect, sign in to AIBIMTool and allow access. Then ask Claude to design a building.

Run it locally

The same package is an MCP server when run with no arguments, so Claude Code, Cursor and Windsurf can drive the API with an API key and write IFC files to disk. Drop this into your client's MCP config; VS Code wants servers rather than mcpServers.

1{
2 "mcpServers": {
3 "aibimtool": {
4 "command": "npx",
5 "args": ["-y", "aibimtool-mcp"],
6 "env": { "AIBIMTOOL_API_KEY": "aibim_sk_xxxxxxxxxxxxxxxx" }
7 }
8 }
9}
1{
2 "servers": {
3 "aibimtool": {
4 "command": "npx",
5 "args": ["-y", "aibimtool-mcp"],
6 "env": { "AIBIMTOOL_API_KEY": "aibim_sk_xxxxxxxxxxxxxxxx" }
7 }
8 }
9}

Tools it exposes

  • create_buildingMetered

    Starts a new design from a brief in the user's words. The agent researches the building type, may ask a few questions or propose a plan, then models storeys, walls, slabs, doors, windows, rooms, stairs and the roof. Returns a project id at once.

  • get_project_statusFree

    What the project is doing and waiting for, the agent's latest reply, its questions with their options, its plan, and the element count. Waits up to 50 seconds while the agent runs, so an agent polls in a loop without burning turns.

  • send_messageMetered

    The next message to the project's agent: the user's answers to its questions, a change to the model, or a follow-up. Runs in the background like create_building.

  • approve_planMetered

    The go-ahead for the plan the agent proposed for a large brief, once the user has seen it. The agent then builds every task.

  • continue_buildingMetered

    Goes on where a long build stopped at its time limit. Long builds normally carry on by themselves; this is the fallback.

  • get_model_imageFree

    A picture of the current model: a 3D view, or the plan of one storey with every room named and measured, so the agent can check its work and show it in the chat.

  • get_ifc_downloadFree

    The model as an IFC4 file for Revit, ArchiCAD, BlenderBIM or Tekla: written to disk by the local server, a seven-day link from the hosted one.

  • list_projectsFree

    The account's projects with their ids, so a change goes to the building the user means rather than a new one.

  • stop_buildingFree

    Stops a running agent after the step it is on. The steps it took are charged and the model keeps what they built.

  • check_creditsFree

    Plan and credits left, so an agent can check before starting a batch instead of failing halfway through one.

A building takes minutes, longer than a chat holds a tool call open, so create_building returns at once and get_project_status waits while the agent works. The agent's questions and plan come back into your chat, so you decide, not the model. A message is charged for the AI it uses: 1 credit per $0.02, at most 150 credits per 8-minute segment, and a run that fails on our side is refunded in full.

Skills

The aibimtool-skills repository teaches an agent what the schemas cannot: how to write a brief the agent builds well, when to ask you, how to handle its plan, what to do with the IFC. Also a Claude Code plugin with the server built in.

npx skills add MeltFlexDevs/aibimtool-skills