What is the problem?
Got mail from Make.com stating that Claude now was able to create Make Scenarios due to new Make-skills.
Bought Claude licence, Installed all new Make-skills, worked daily 3-4 hours for a month trying to create a “upload more documents” function to my flow, before final analyze is generated. Claude keept making faulty blueprints that did not work, so i scrapped that plan :-/
What have you tried so far?
Everything, different claude versions, Claude code etc.
Please 
Just like the make.com AI agents, Claude for make, infact Claude for almost everything is still at the Beta stage which means still undergoing testings. Claude for make might add modules to your workflow or plan your workflow but you are the pilot. This tool infact almost all other tools are co-pilots not autopilots.
Plus, I must be honest. It takes some quality time and interest to get really proficient at using these tools. Claude isn’t a genie that has it all figured out.
Don’t generate Blueprints please. Agentic Claude uses hands (tools to do actual building).
Generating a blueprint would mean you trust that clause has all updated make.com modules json schema in its training library which is quite impossible. Using agentic Claude on make means the Claude agent can access your make.com and can plan workflows, search and import modules, write filters and formulas, map data etc.
So you’re on the right track if you need a little help I’d be glad to assist.
Hi @Martin_Gosvig , thank you for sharing your experience! This is exactly the kind of feedback that helps us improve.
We’re sorry to hear that the blueprint generation didn’t work out as expected. For the best results with Make Skills, we currently recommend using flagship AI models. Currently, Claude Opus 4.8 is the strongest option and tends to handle complex blueprint creation most reliably.
We’re actively working on improving Make Skills based on real user feedback like yours. If you’d like to share more any more details, we’d genuinely appreciate it. Here’s a direct feedback form: Make Skills for Claude
In the meantime, we’re also working on more resources and guides on how to create Make automation scenarios with Make Skills for Claude, so stay tuned!
Hope to turn this into a better experience for you!
I use Claude Code + Make daily (VS Code) to generate and create full automation flows. Here are some of my experiences that may help:
-
In general, when claude context gets bloated, it is more prone to errors, so manage your claude context window well with /compact and/or /clear when needed.
-
I maintain an active project status and context document locally for each product, including a register of pervious bug fixes and then only let Claude and Codex GREP for relevant sections and status so in addition to 1) above, I manage the context as best I can.
-
Don’t let Claude do all the debugging, it does not have access to the execution log in great detail, so may waste your time. Inspect any errors yourself and provide this specifically to Claude.
-
Always check each module if Claude/ Codex added a module - I find that with many Make modules, the omit some settings and a small omission like that can trip everything. In my case usually the HTTP modules and required “METHOD” is usually omitted as example.
-
Don’t rely on Claude to always pick the right modules. - Be very specific in your prompting and design and give Claude guidance on the modules you would prefer and plan with Claude code first, before deploying. You can also use ChatGPT if you have a paid account in the WebUI and connect Make MCP as a connector that can help you build and plan together with Claude or Claude code, this way you can split the creation logic and planning between the two of them. You could even use free ChatGPT or similar for general planning without MCP connectors.
Paste a screenshot of your flow or part thereof and where you are getting stuck?
Also are you using the Make MCP, CLI or API?
Where are you using CLaude? Web, co-work or Code locally or browser?
Hi and thanks for your input.
Since this is my first time building a website and setting up the backend, I’ve used everything from Gemini, paid ChatGPT, and Codex, to paid Claude and Claude Code.
My original workflow actually worked great. It allowed customers to upload documents and subsequently receive an analysis via email.
When I wanted to improve the setup by letting customers upload an additional document—which an OpenAI module needed to provide an even better analysis—I tried using Claude and Claude Code.
I don’t have enough technical knowledge to catch Claude’s mistakes myself, so I just had to trust that it knew what it was doing. It didn’t, and I eventually had to give up after spending 100+ frustrating hours trying to fix it. What a bummer!
I also wanted to connect a legal precedent API (domsdatabase) to the customer analysis, so similar cases and rulings would appear in the report. However, I’ve had to drop that too, as neither Claude nor Codex can figure out how to do this in Make. So frustrating!
Ok, what are you using for the website? What platform are you deploying on? I can relate to your use case as I recently created a legal system with Make (actually submitted it to the make Community challenge)
Again without clear idea of the current modules you are using for document intake and the process flow, it will be difficult to navigate the exact issue you are facing as it may be a MAKE blocker or it may be a blocker on your website flow. ( Upload more documents path may be disconnected)
Thanks. I am using Tally for the intake form.
However, I have completely redesigned the architecture now. I stripped away all loops, database integrations, and complex multi-document paths. The workflow is now a 100% linear conveyor belt: Tally (text/image) ➔ Gatekeeper ➔ Router ➔ 1 Specialist (gpt-4o-mini JSON) ➔ Gmail.
By keeping it strictly linear, I avoided the Make blockers entirely. I am currently just finalizing the prompts for the individual specialists.
Awesome. I could get the make-skills for claude added to my organisation settings and ask my team to use the skills. Of course overtime, we will finetune the skills to our more-specific needs. Thanks make.com
We have also considered the make-skills and fine-tuned it as per our need. Having the make-skills were extremely useful as a head-start. So it might also be quite useful if we finetune the skills as per our respective needs.