Web2.0 strikes back: Using MCP to make your life easier

In my previous post, I've mentioned briefly that I think that the place of a LLM in my design workflow slowly evolved into the thing that I can slowly, quietly, offloading the annoying parts to.

Since then, that process has accelerated to something more structural: integrating AI tooling – and especially the MCPs – directly into my everyday workflow. The mental shift has been from asking "What can I prompt this thing to do?" to giving the model enough context to be able to improve my efficiency at everyday things that I have to do anyway.

MCP: the new web2.0 (for now)

I still remember the height of the Web 2.0, where the promise was a "mashup" culture – pulling data from one tool to another, blending feeds, and creating new value by connecting the dots.

And then the reality struck. The APIs we relied on either got closed down entirely or their owners put exorbitant price tags on accessing them. As often is the case, the open data culture of the web got hamstrung by corporations trying to squeeze more profit out of something they offered for free.

MCP is this, again (for now). It allows me to plug multiple data sources as context into my LLM of choice and use the LLM as a mix-and-masher of this context.

My setup uses four MCP connections that have fundamentally changed how quickly I can get to "shitty first draft":

1. Notion MCP – Pointing a model to our existing PRD templates, past project docs, and all sorts of documentations and guidelines to populate context window. This allows the model to write in our actual format, use our specific terminology, and can even pull in relevant sections from previous, similar projects as reference.
2. Atlassian MCP (Jira) – instead of me manually trawling through backlogs, I can ask the model "review existing product ideas and bugs on <feature>" and get a somewhat organized and prioritized list back. It reads the actual tickets, synthesizes the themes, and links directly to the relevant Jira tickets.
3. Figma MCP – pretty obvious. Once I have the design in place, it allows connecting to existing Figma mockups and prototypes and build out working prototypes, pulling in actual variables and components as it goes. This goes hand-in-hand with the fourth one, which is
4. Storybook MCP – the synergy between what's designed in Figma and what's actually available for engineering helps cut down on hand-off, explanation and back-and-forth and allows us to get to the solution quicker. I never enjoyed the disconnect between "blueprint" (mocks) and "material" (code), and Storybook MCP with code generation does a pretty good job bridging this gap.

The model with the MCP attachments acts as a counterpart that has already done the reconnaissance across the entire product stack and it enables a rapid, iterative dialogue between these tools that was previously impossible.

The value is in the continuity. The insight from a note in Notion taken during an interview connects to a year-old Jira ticket, which informs a PRD draft written using a proper template, which translates into a prototype using real components from Storybook. It's pretty fun and I would recommend trying it if you haven't yet.