Loading video...
Video Failed to Load
Introducing Inspo. A design MCP server for Claude Code, Codex, and OpenCode. It searches 800+ beautiful websites and finds relevant design inspiration for your coding agent. Install → npx inspo-mcp install
148,037 views • 1 day ago •via X (Twitter)
48 Comments

Check out Inspo MCP below! It's free & 100% open source.

This will be big

🙏

Amazeee This is like mobbin but opensource I'll try it as soon as I get on my PC

Yep, would love to hear your feedback after you try it!

Will try it out once I get home

lemme know what you think!

this is dope

thanks man!

huge 🔥

ohh man, i got an lifetime plan for mobbin just to use the MCP lol 🫠

Bookmarked! I'll give it a try!

Would love your feedback if you end up trying it out!

Let's go!

Fantastic work on this!

Great work

Thank you!

This sounds useful for the “I know it when I see it” stage 😅 Does the agent get screenshots too?

Yep!

The gap this closes is real: agents write clean code and then default to the same three layouts because they have no visual reference to pull from. Curious how the 800 sites get ranked for relevance, by category or by prompt embedding.

800+ sites for design inspiration? My coding agents are about to get a serious style upgrade. Finally, no more ugly UIs by default.

Definitely trying this out. Much needed.

I tried npx inspo-mcp on a Claude Code UI pass — design refs stuck better than dumping screenshots into chat.

I think it is funny that even with all this 800+ websites, the results you shown in the video are EXACTLY what my claude code did with the last 3 websites I built. I have a feeling about design and AIs... 😂

Totally fair feedback! One thing I really like about it is it asks you a few questions and forces you to clarify what you're building, then suggests some relevant UIs. It also works especially well for open models that are cheaper, so you can iterate even faster

This is quite cool! I'm curious, what was the biggest thing you learned from building this? ps. gave you a star on github. pps. love your "build first, ask permission later" ;) approach. very cool tool and smart design, clearly.

Goat strikes again

🙏🙏

The 800-site index is the interesting bit here. How do you rank relevance across such different design systems?

love this!

thanks man!

Design MCP servers make retrieval part of coding, but the interesting contract is provenance: why a site was selected, what constraints were inferred, and how an agent cites or reuses it. alibaba/page-agent is a useful adjacent reference for turning UI context into actions.

I can just tell that this will be massive

800+ sites is a solid dataset. Curious how it handles semantic search for specific UI patterns.

这个思路挺实用,省得每次都从零找参考。

websites only? or... ios apps too? o'm sorely lacking in app inspo

websites & web apps only for now! I haven't gotten into ios apps tbh, you build ios apps? 👀

I am looking into my first one

这个方向挺实用的,agent 终于不只是在仓库里翻文件了。要是还能把“为什么这个页面值得参考”一起结构化出来,生成结果会比塞一堆链接更好用。

Clever move — it shifts design taste from prompt-craft into retrieval. Curious how the agent handles conflicting inspiration: does it blend top results, or pick one reference and commit? That choice is where output quality usually breaks.

This looks like a game-changer for finding exactly what I need for my next project! So excited to try it out. ✨

What I find fascinating is that we're optimizing AI to replace tasks that most engineers already automated with a bash script 10 years ago. The real frontier is AI handling the truly ambiguous decisions.

Design inspiration as an MCP tool is the right layer — agents ship ugly UIs because they lack taste references, not code skill. Searching 800 sites beats prompting "make it look modern." Do you rank results by visual similarity to the current page, or by component type?

800+ sites is enough to dodge generic AI slop. MCP keeps design search as a tool call, not prompt soup. Latency vs relevance will be the real test.

This is a sharp MCP use case because it turns taste into a callable capability, not just a link dump. The 800-site corpus is useful only if the ranking stays opinionated. What signals are you using to decide which examples belong near the top?

it looks ai, users would rather to do what they are doing now thanks

Agents still spit out the same generic card layout. Dumping live sites into context is crude but it will beat another “make it clean” prompt.

Fun fact I am looking for oss of mobin mcp it is going to huge


