Graham Lipsman

I wrote a Cursor<>Figma MCP server to make it easier for AI copilots to implement designs. It's not perfect, but the resulting HTML and CSS is a lot cleaner than giving AI a screenshot and having it guess. Link in the replies, take it for a spin and let me know what you think!

Hooray! Thanks @trq212

@threepointone I don't own enough NET

iykyk

@lymanstoneky I'm surprised nobody has mentioned /sandbox Doesn't get rid of all permission requests but it has FAR fewer, and it gives me better piece of mind than dangerously skip permissions

Claude knowers: how do I make Claude stop asking me to approve a billion little things? How do I make it just do the stuff in the plan it developed and stop asking my permission?

@ctjlewis Wouldn't their failure imply the social contract is still alive?

Should wait until the poll’s done but I can’t resist. If you look at the Cluely saga and your high level takeaway is “startups are hard,” I’ve got a bridge to sell you.

Asking Claude to provide ideas based on "the existing literature" is OP

@thisismahmoud @yminsky Woohoo!

@glipsman @yminsky We got an official Railway Agent skill. Ping me if y'all got any questions or if there's anything I can help with 🫡 github.com/railwayapp/rai…

So, where is the application hosting platform of the future that's optimized for vibe coding? There are a ton of little applications I'd love to build for myself if the costs and annoyances of setting up services and permissions were mitigated.

@dillon_mulroy Someone's been playing with RLMs

subagents are sparkling map reduce

@dhasandev Well yeah but the same policy applies to the OP's post

@glipsman yep but not for ambitious use cases unfortunately

@dhasandev Agent SDK supports oauth for local use, you can use your Max plan—officially sanctioned x.com/trq212/status/…

@glipsman oauth subscriptions are better

@mattpocockuk Once it can use my MCPs it'll be golden. Flies blind without eg comtext7

docker sandbox run claude .

@tokengobbler Though I don't use Linear it seems like their AI strategy has been pretty solid so far.

When your AI strategy is a dropdown menu

@_brentbaum Very cool! I'd love to try using this data to preserve flow states—e.g. block certain sites or put my phone on DND for some capped amount of time

the crazy thing is that it's actually working

@trq212 Was using memory with subagents, gets outdated so fast. Maybe I'm asking it to remember the wrong class of things? Any tips?

We've rolled out a new auto-memory feature. Claude now remembers what it learns across sessions — your project context, debugging patterns, preferred approaches — and recalls it later without you having to write anything down.

@tannerlinsley @amsryq Yes, lock-in is the least of Next's problems

@amsryq You still made the right call.

This after I just migrated my project from Next.js to TanStack Start 💔

@stephenhaney This is it—congrats on the launch and imminent liftoff! 🚀

Hello! Today we're releasing Paper Desktop Paper is now a canvas for Cursor, Claude Code, Codex. Any agent can read and write html to Paper. • push or pull from your codebase • pull real data from anywhere • less work, more design What will you ship? Sound on 🎶

I put a skill together for setting up a bunch of lint rules, inspired by the OAI harness engineering post. I'll write a better README at some point but you can just ask Claude to explain it for now. >30 rules for Tanstack Start, Drizzle, and Biome setups github.com/GLips/enforced…

@_ashleypeacock No, but I should make it so. I built it with Anthropic's Agent SDK but under the hood it's the same code as a downloadable MCP would be. Easy enough to extract

@glipsman Really cool! Is it a public MCP?

I was thinking about MCP and Code Mode this morning, and I got thinking how cool would it be if you could convert any MCP server to use Code Mode? This reduces token use of any MCP server, which is nice, but if you need 10 MCP servers, you quickly bloat the context again. So then I thought, why not merge all the MCP servers together behind a single URL and have a single search/execute tool for all the MCP servers your agent uses? I then realised, by re-reading the Code Mode blog post, that the clever people at Cloudflare had already started thinking about this, and I'd totally forgotten that Cloudflare MCP Server Portals already exist. In effect, it lets you compose multiple MCP servers together behind a single URL, with unified auth and access control. Throw Code Mode in there, and you've got yourself a real winner that solves everything I was thinking about this morning. I hadn't started using MCP Server Portals purely because it was mostly a "nice" enhancement, but definitely nothing ground-breaking, but once Code Mode is enabled, I think we'll be seeing a lot more MCP Server Portals! Anyone used MCP Server Portals yet, or given Code Mode a go?