SFO · San Francisco
LicaWorld MCP
Redesigned Lica's MCP so non-technical creators — PMs, marketers, designers — can recruit and chat with a team of agents the same way they'd talk to one AI.
- Role
- Product Designer (Contract)
- Timeline
- NOV 2025
- Technical Dialect
- Figma
Problem
Canva and ChatGPT have trained people to talk to one AI — a single conversational voice. Lica's MCP unlocks something different: a roster of specialist agents working together. But the people it's built for — PMs, marketers, designers without engineering chops — had no familiar way to recruit a team, switch between agents, or scale past five without hitting decision paralysis.
Solution
I designed two flows around the mental model people already trust. In recruitment, users describe what they want to make and the system templates a starting team from pinned domain experts — Reddit for data, Canva for design — so every sector is covered without name-guessing. In chatting, the Lica logo represents the whole team (one voice, like ChatGPT); pinned agents sit behind it in black-and-white to show they're still in the room. Picking one promotes it to full color, replaces the logo, and rewrites the prompt to match its domain. Same flow works for five agents or a hundred.
§ Shipped
What I built
The Process
Design architecture
Optional deep dive — how the problem became the shipped product.
Mapping the mental models people already trust
I studied how PMs, marketers, and designers actually use Canva, ChatGPT, and Notion AI — and found one shared habit: people expect to talk to one comprehensive voice, not orchestrate a roster. The UX gap I designed against: Lica's MCP gives them a roster, but no familiar way to recruit, swap between, or scale it. I put myself in the user's shoes — if I had a hundred agents, I'd query the same handful repeatedly and ignore the rest. That insight drove the whole interaction model.
Recruiting a team without naming agents
Users don't know what their agents are called — and shouldn't have to. Instead of a search-and-pick gallery, I built a prompt-first recruitment flow: describe what you're making, and the system templates a starting team from pinned domain experts (Reddit for data, Canva for design, etc.) so every sector is represented. Pinning the most-used agents handles the scale problem — five or a hundred, you still pick from a handful, and the system covers the long tail. The user feels empowered, not paralyzed.
One voice on the surface, the full team underneath
The Lica logo stands in for the whole team — mirroring the single-voice pattern from ChatGPT and Canva so cognitive load stays low. Pinned agents tuck behind the logo in black-and-white to show they're still contributing without being directly addressed. Picking one promotes it to full color, replaces the logo, and rewrites the prompt placeholder to match its domain — pick Reddit and the field becomes 'let's make some data-driven decisions.' The handoff is visible, but it never breaks the conversation.