Looking for a bit of guidance please. I am sure there is a question like this, but not found the answer yet. I successfully connected Claude to Pyrevit MCP in Pyrevit 6. This is fantastic, not only reading my revit model, but letting me carry out repetitive tasks like duplicating sheets and renaming etc. I could do this in Dynamo, but I do these tasks quite infrequently and I always get some node wrong, plus I cannot write Python.
So, the drawback was that I don’t pay for Claude Desktop and was looking for a free tier agent to do this in Pyrevit 7. I also do have Google Workspace, so tried first to use Gemini as the agent.
I was using Gemini to guide me on setup. It got me to setup Gemini API, to get a key. Then install Visual Studio Code, where it set up an extension called Continue. This was supposed to use the Gemini model to talk to Revit. Eventually, I got it working BUT, using Gemini Flash 3.8, the servers were always busy and I didn’t have Pro credits.
I ditched this approach and asked Gemini for a new method, and it got me to install Ollama, then use Continue Ollama provider and qwen2.5 as the mode to search. I can’t seem to get this to work and Gemini hasn’t been very helpful in assisting me.
Am I approaching this all wrong ? Is there a simpler way ? Is there a recommended agent and guide available ? Would really appreciate your help. Thank you.
My personal opinion. If you’re going to use it - use it. A lot. And for that a paid account is probably worth it. You can use it for much more than just Revit of course. I wrote a spec last week without opening Word. Sop for $20 (or $17) it is possible to get value out of AI.
Or your best value is to ditch dynamo and bit the bullet and learn some python. That paired with AI is far far more powerful than just running a MCP.
Unfortunately AI agents use compute, and compute costs money. Google, Anthropic, OpenAI, etc, all have fee teirs, but if you’re using up your token limit with Claude you’ll likely hit the same limits with the other cloud models.
I’m less familiar with Gemini but you should be able to connect to pyRevit with one of their desktop apps without needing to use VS Code and API keys if you want to use Gemini because of your Google Workplace subscription.
I’d stay away from trying to run a model locally, like qwen. That’s a complex process and your computer likely can’t run a model large enough to drive pyRevit nearly as well as Claude can.
In a year or two we may all be using free local models, but for now I’d just pick a paid plan with whichever company you prefer.
Thank you both. I have done a few good tasks so far still using Claude desktop. For example, I duplicated 42 views, renamed them, applied the right parameters for my browser organisation, then applied view templates. A relatively simple task, but time saving. Didn’t run out of credits there.
It looks like there is a Gemini pay as you go version, where this sort of computation would only add up to about 1p of cost. So you could easily do 20+ tasks a day for £1 per week. And I won’t be doing anywhere near that range. Well not at the moment anyways, just a simple Arch Technician/BIM Manager.
Opencode has a bunch of decent free models if the cost is the main issue. I extensively used space bunny free to get V7 out. And while I could see the seems with better models, it did and still does a bloody great job. Before the z.ai model was stealth and free and kicked ass.
Claude, or ChatGPT are very good value for the money especially taking lower tier model (Luna and terra or sonnet are good enough for mcp tasks) and they are smarter when it comes to MCP and setup and coding
Roman did a great job with the new built-in MCP on the setup side. Check it out.