Click Reply to post your comments about Figma, Inc. (FIG:USA).
For our latest product enhancement (the screener report which is truly awful now) our UX person designed it all painstakingly on Figma. The end result is an enormous canvas with some 70+ screenshots all connected with state transitions and labels. All very complex and a hard thing to digest for the UX approval team, hard to follow the different versions being proposed, and most importantly, to judge which one "feels right". Seriously, just look at it:
But then the magic happened.
Via Figma's excellent MCP server the entire design was converted to a functional prototype in a matter of minutes. Several changes were decided purely by using the prototype and communicated back to Claude which updated both the Figma design via MCP with precise PRDs, and the prototype. I have not even tried to follow the design specs on Figma.
This is the first example I've seen of AI integration that makes a complex product a heck of a lot better. This is exactly what we want Alphanaut to become, and we're are trying to learn from Figma's example. No other AI integrations have impressed me. For example "Breeze" agent on Hubspot is annoying, and so is "Rovo" on Atlassian Jira. Perhaps I'm misusing them but what we achieved with Figma is the only a-ha moment I've had with AI outside of using LLMs directly.
Needless to say I think FIG is rock bottom cheap SaaS. I'm buying all I can. And earnings are around the corner.
Have you experienced this with other complex products? Who has greatly improved their user facing product with AI?
PS. other companies that claim AI has improved their operations internally. For example Lemonade claims AI has made them much better at the insurance business: underwriting , risk assessment, etc. Those AI claims can only be judged when they report their earnings.
Why not follow Figma's example and create an MCP for P123? Personally I would rather not have to use Alphanaut when I could do it with Claude. Is Alphanaut good at coding?
If connecting Claude to an MCP server made your workflow that much better with Figma, why shouldn't Portfolio123 do the same for its subscribers?
You can wrap the P123 API into a custom MCP server in about 45 minutes. I have done this, timed it, and Claude wrote it in 45 minutes.
Once that bridge is up, Claude Desktop or Claude Code can interact directly with P123's engine through natural language—pulling factor ranks, running screens, and handing off data directly to local Python scripts for custom quant analysis without ever touching the browser UI.
To be clear, the real value comes from Claude being able to manipulate the data with local Python code. The primary use case is letting Claude fulfill all the custom feature requests P123 doesn't have the development bandwidth to build.
For example: if I want to run a K-nearest neighbors model on ranking outputs—something P123 will probably never build natively into its AI toolset—Claude can write the script, pull the data through the MCP, and execute it locally. And it runs entirely on my own compute.
I've already done this using the API and the MCP Claude created. Because of that workflow, I don't actually need P123 to keep building endless new UI features—the platform's value becomes the indispensable data and ranking engine behind the agent.
1 Like
100%. By "Alphanaut" I meant our overall AI approach
We have an internal MCP right now. Making it public as a simple wrapper of our API is easy as you say. Then we simply focus on improving our API/MCP so you can unlock the magic instead on endless UI features which are cumbersome to use for true data mining anyway. Unless they are purpose build data mining features, which never satisfies everyone.
3 Likes