Can We Use Claude Sonnet with Bonsai Using MCP ?
Hi everyone,
This is my first post here—I’m new to Bonsai in Blender and still getting familiar with the Open IFC shell. Recently, I saw a video about connecting Claude Sonnet (via MCP) to Blender, and it made me wonder: could we use this with Bonsai too?
https://github.com/ahujasid/blender-mcp
I think it could be a game-changer for working with IFC files, like:
- Editing IFC data more intuitively using natural language.
- Validating IFC files for errors or inconsistencies.
- Remapping property values or attributes automatically.
Does anyone think this is feasible? Could this combination help streamline IFC workflows?
Also, if you have any tips for a beginner on Bonsai and Open IFC, I’d appreciate the help!
Thanks,
Tagged:
Comments
We've had a few discussions about AI here, but I agree now it's a good time to talk about it again.
My LinkedIn feed has been overflowing with these AEC ai posts for a few months now and they all have one thing in common: they try to sell a solution they have, for problems nobody has. Or in other words, what we are all looking for are use cases for ai.
Let's go through the use cases you and others suggested and see if we can find something useful.
1. Editing ifc (or more generally bim) data with natural language: this covers a huge range of tasks. In short: concept design is possible, but this is not something many architects want to give up. Turning your concept into full ifc model is possible with homemaker without ai. Autonomously designing whole projects is impossible right now. Detailed design is possible , but much faster per hand.
2. Validating BIM data: this makes no sense. Validation needs to be 100% reliable and therefore clearly algorithmic.
3. Remapping values in BIM model: also highly problematic. In simple cases writing a short algorithm is much faster than trying to use AI. In complex cases you will never know if the job was done correctly. How do you deal with this?
4. Visualization: there are tons of awesome ai possibilities, we should definitely document some open source options!
2.a support with definition of standards for validation: I believe AI could be well used to go through existing human readable standards and translating them in computer readable rules. But this basically needs to be done once and checked by a human, so I'm not sure it's worth it.
5. Data extraction from BIM models: I suppose there is still a lot of people who only need to extract data and don't have capacity to deal with simple computer readable filters. (Investors, cost estimators?) The main issue here is reliability, but again, you could just use LLM to translate your human input into the computer readable filter.
6. Analysis of point clouds: this should help immensely, if anyone knows an os solution that divides point clouds in separate objects that would be awesome!
7. Analyzing contracts, scopes of service: this is already working and many people use this (analyze this contract and give me a schedule of rooms that need to be planned)
8. Automatic dimensions: I've seen this marketed as AI usecase, but personally don't believe it can work better than an algorithm. I do believe we should provide an algorithm for automatic dimensioning, though.
So generally right now I believe anything that uneeds simple scripts or filters or prompts could use AI well and we should think of integrating it in Bonsai!
@JanF
I think I fall in this category (Quantity Surveyor), extracting IFC data in Bonsai, with some little training, is easier than using formulas in a worksheet in Excel, pivot table or similar
I wouldn't hire anyone who needs AI to carry out such a task ;)
@JanF
This describe exactly what (many times) i think about AI. Thanks for sharing it! :-)
@JanF > @JanF said:
Florent Poux is expert in this. Here's an easy to follow tutorial, all python:
3D Clustering Mastery: How to Segment Point Clouds with Graph Theory
Love this! After analysis like this, automatic conversion to an usable editable ifc model should be also doable!
And to stay on topic, AI could be used on the analysed model to classify the objects. Not sure for what purpose though.
Hey! I've been tinkering with this, it's definitely doable. The main thing I've been doing is setting up commands in Blender using the Python IfcOpenShell commands and then making an equivalent tool for Claude to call. It did work but I'm heavily focused on my own use case, which is a bit away from the standard Architectre use cases. I'm really interested in discussing this further with someone who has other use cases in mind. While I agree with janF's opinions, my own is that there are infinite cases where these LLMs can be of service to the industry.
I think about this for specific tasks that require a lot of effort for modeling.
This discussion is proposing an approach for a door and window modeling tool. Due to the complexity and number of parameters involved in this work, in addition to the need for freedom for designers, I think that a model trained specifically on door and window data, feeding an agent for modeling following specific rules for this purpose, might be an approach that would work.
Another use case along the same lines of thought would be the modeling of steel reinforcement for concrete. Modeling a bar is simple (a swept solid), but doing this for dozens of parts of various diameters and different shapes and then putting everything together in an assembly is complex and laborious. An agent that could understand a geometry (e.g. a column) and a structural calculation memory for this element and suggest the necessary bars.
I don't have the capacity to develop this, but I would like to see someone try... ;)
An example of what's to come: AI agents for civil engineers