Using AI Agents to assist in modeling : Honeybee-MCP(beta)

share
1

Hello, everyone!

I created a Honeybee-MCP to link HBJSON with AI Agent programs, which enhances the ability to perform fuzzy editing on existing models. This project is built entirely on top of FastMCP and the Ladybug Tools SDK.

LoftyTao/Honeybee-MCP: MCP Tools by LBT-Honeybee

While I acknowledge I may not be the first to do this, I believe further engineering of the MCP warrants more discussion.For example, under what circumstances might certain commands be issued. Perhaps we could convert these into engineering-specific prompts to guide the AI.

You can install Honeybee-MCP in any AI IDE to access its core features, such as Cursor, OpenCode, and VS Code.

Currently, this MCP can handle some basic scenarios, such as:

  • Loading and saving models
  • Querying models or objects
  • Removing features like apertures, doors, and shades from models, rooms, and faces
  • Generating apertures proportionally for specific surfaces
  • Generating louvers for specific surfaces
  • Other features that may be developed later…

In the tested cases, I issued commands such as:

Load the revit_sample.hbjson model and filter for all rooms defined as “Bedroom” to identify their south-facing exterior walls; subsequently, remove all existing windows on these specific walls and generate new rectangular apertures with a 65% Window-to-Wall Ratio (WWR), maintaining a 0.6m sill height and a 2m window height; finally, configure exterior blinds for these new windows, applying randomized parameters (such as slat angles) to simulate realistic usage scenarios, and save the modified model file.

Finally, I only need to keep acknowledging the MCP command requests to complete this case:

image
image1211Ă—626 17.5 KB

:grinning:This is a summary of the AI Agent’s automated operation. I can’t even finish a cup of coffee.

## Summary
Model Loaded: Revit_Sample.hbjson

- 15 rooms total
- 295.74 m² floor area
Bedroom Rooms Identified: 3 rooms found

- Bedroom_1_fd751b47
- Bedroom_2_614a5393
- Master_Bedroom_d6e36276
South-Facing Exterior Walls Identified: 3 walls

- Bedroom_1_fd751b47..Face8 (azimuth: 180°)
- Bedroom_2_614a5393..Face8 (azimuth: 180°)
- Master_Bedroom_d6e36276..Face5 (azimuth: 180°)
Modifications Completed:

1. Removed Existing Windows: 1 aperture removed from each of the 3 walls (3 total)
2. Created New Windows: 1 new rectangular aperture per wall with:
   
   - 65% Window-to-Wall Ratio (WWR)
   - 0.6m sill height
   - 2m window height
3. Added Exterior Blinds: Configured with randomized parameters for each bedroom:
   
   - Bedroom 1: 4 louvers, 0.45m depth, 45° angle
   - Bedroom 2: 5 louvers, 0.35m depth, 60° angle
   - Master Bedroom: 3 louvers, 0.55m depth, 30° angle
Model Saved: Revit_Sample_Bedroom_South_Walls_Modified.hbjson

The exterior blinds have been configured with varying louver counts, depths, and angles to simulate realistic usage scenarios across the different bedroom spaces.

The primary constraint of the project at this stage is the limited diversity of practical use cases. I actively welcome constructive feedback and feature requests from the community, which will be prioritized for future implementation.

To enhance the project’s utility and integration within building performance workflows, the following features are planned for future releases:

  • Implementation of HTML model exporters.

  • Development of “Apply” utilities to streamline the batch assignment of Construction Sets, Program Types and HVAC Templates.

  • Add Agent Skills and Engineering Sample Prompts

Now you can view the tutorial (with English subtitles) on Youtube.

bset,
Zhengrong

8 Likes
2

Hi @ZhengrongTao . Super excited with this new features with LBt. Can we have a tutorial video on how to install the mcp for lbt?

3

hi,@Asisnath

Since my native language is Chinese, I may have Chinese-language demo videos and user analytics for my channel. However, for English videos, I might use AI voiceovers. If any releases become possible, I’ll respond to you again.

Additionally, here’s my suggestion: MCP is likely very familiar with LMM tools. You can use any AI IDE to ask how to pull and build the contents of my repository, then let them handle the automatic configuration. I suspect this process should proceed without any obstacles.

best.
Zhengrong

2 Likes
5

This is fantastic. i can see this being very useful towards the end of concept and early prelim where a fairly resolved model exists and some “what - ifs?” need to be tested. It offers a more faithful test of an architects design than a parametric analytical model but offers a similar level of speed.

1 Like
6

hi,@Asisnath

I’m currently planning to create a Chinese video tutorial, but I still can’t determine the schedule for making an English video tutorial.

I wrote a simple demonstration document for Honeybee-MCP, deploying and installing all the configurations and dependencies (except the project files) using OpenCode.You can either have OpenCode review the README.md, or use the Prompts provided in the demonstration document to enable the AI Agent to perform the deployment and testing in an entirely automated manner.

best.
Zhengrong

2 Likes
7

Thanks but a video will help a lot!

8

hi,@Asisnath

Now you can view the tutorial (with English subtitles) on Youtube.

I recorded a simple video. Due to other tasks to be done, I had to use English subtitles to edit the video that I had already sent to the users of the Chinese community and then upload it to Youtube. Fortunately, the main documents are in English, so I think there shouldn’t be any major problems.

best.
Zhengrong

3 Likes
9

Thanks, can you share the link please to the tutorial?

10

Hi all,

Now, Honeybee-MCP provides the HB-MCP Writer and HB-MCP Reader components for connecting with AI CLI tools (such as OpenCode CLI, Clawde Code CLI, etc.), enabling the execution of simple or complex model editing through natural language, and this process does not require leaving the Grasshopper interface.

This can prevent the kind of chaos that occurred in the past when editing, deleting and setting certain objects in complex Grasshopper components and data structures.

Just like in my current case, in previous cases, we needed to go to the source of the model, find the certain geometric object that needed to be deleted, and then remove it from the data list. But now, we just need to tell the AI

Remove the exterior window on the east side of the current model

and then have a cup of coffee while waiting for it to report back to you.

testgif8

As for what HB-MCP can do, you can check the feature list of this project to find out.

:robot:Since the development of this project requires active feedback from the general public, I sincerely hope that everyone will enjoy using HB-MCP and share their experiences with me.

1 Like