magic dima

LOG 7, day 169 building Faven Set aside the MCP demo video. Announced it with a simple tweet instead. Removed the waitlist requirement for closed beta access. Now we are in open beta. Anyone can now try Faven and get paid upfront for their limit orders. Caught a signal from @vibhu about Farcaster changes that look promising for Solana projects. Shipped today a Farcaster mini-app landing + channel. Full app is ready for launch tomorrow. Come say hi if you’re on Farcaster farcaster.xyz/~/channel/fave… In get-things-done mode.

LOG 5, day 167 building Faven Video that was supposed to announce and demonstrate Faven MCP got pushed back one day. Wrote the script, recorded the whole thing, and right at the end the MCP server hit a similar premium calculation bug to the one I had last week. Take is useless. That’s the WTF of the day. Video goes out tomorrow. Bug is already fixed, and MCP picked up 4–5 small but very neat improvements along the way.

LOG 4, day 166 building Faven MCP is done. The landing/onboarding page for MCP is done And the onboarding prompt is done. I had planned a skill for Faven, but I simplified its implementation: instead of a full-fledged skill, I created a "prompt"—essentially the same thing, but presented as a single file whose contents you simply need to copy. Everything is live. Any MCP-compatible agent — Claude, ChatGPT, Grok, Hermes, OpenClaw, and others — should now be able to find its way and onboard itself to Faven. Going to publish a video with MCP tomorrow. WTF of the day. Spent an hour+ debugging a broken tool that runs Solana programs on my laptop so we can actually develop stuff with comfort. One hour doesn’t sound like much, but in an hour these day a dev can shipping a medium-sized feature. (Normal tweet ends here. Nerd content about Solana local development follows till the end.) Here’s what happened. Faven stopped working on my laptop yesterday evening. This morning I sat down to fix it. Transactions kept failing with “account not initialized.” But the account was initialized. Turns out the recent rent reduction on Solana works fine on mainnet and is completely ignored when you’re developing locally with Surfpool. @surfpool_sol — that cool tool I use to develop Solana programs on my machine. Surfpool forks (clones) mainnet state, which is why it’s incredibly useful. It copies wallets, accounts, tokens, programs everything — including the accounts Faven requires to function. The problem. On mainnet after the rent reduction the rent-exempt amount for those accounts that Faven uses is now lower. Locally with Surfpool it’s still using the old rent-exempt rate, which is significantly higher. So the same accounts that are perfectly valid on Solana become broken when I run the Solana fork locally. Transactions fail because the accounts are underfunded. I fixed local account by funding them with old rate. A straightforward fix. I reported it as a potential bug. I wouldn't rule out the possibility that I messed something up on my end.

Starting a log on how I'm building @FavenMarkets — the good, the ugly, every wtf. Follow along if you want the raw journey. Let's go. 🚀 LOG 1, day 162 I started Faven in May. Better late than never to start talking about it. Got kicked out and banned from the Backpack discord for creating an issue about my product and backpack wallet. I will just email them instead. A new team member took one look at Faven and asked "where's the MCP for this?" Boom — shipping an MCP server this week for your favorite agent. Market maker partners asked for a feature so they can manage wallets and funds more flexibly. Shipping it ASAP. I keep onboarding users 1:1 — only marketing I know for a product that just launched. I invite you to try it out — tweet or DM me "I want to try".

LOG 1, day 163 building Faven I'm almost permanently dragged into multitasking on 2-3 things at a time. That's terrible. Choosing the right thing to focus on is the main thing to get right. A team of 3 can only work on a very limited subset of things. Doing more things half-finished or half-quality — might work or might not. Somewhere it’s ok, but not on a core feature of your product. Automation is the thing, but you always have to measure the ratio of time you spend vs. the time you’re going to cut. And automation can go wrong — today and yesterday I burned 2x of my daily token budget in a 30–60 minute span after giving one agent the green light to run multiple sub-agents. Our focus is on MCP. Building a good MCP is interesting. It’s a different kind of API — intent and result oriented. MCPs are a different UX beast. MCPs have a wow effect when done well (or when you’re lucky), otherwise, it’s simply awkward and gets in the way.. First internal tests were like "WOOOW" but it is rough when "you're not lucky". Excited to ship this thing soon, 80% done. And finaly a WTF of the day. A team member recorded a short video for the first time in his life. About Faven of course. He witnessed all the usual perks of being on camera (making wierd mouth noises, unnatural stance/pose, throat that goes dry — that kind of stuff), Nontheles he made the video, spent another half a day editing it, then posted it the next morning — only to see it banned in less than 10 minutes. Facepalm.

LOG 3, day 165 building Faven First “Buy Lower” trade executed and settled. User got paid upfront to agree to buy SPCX at $155. Trade settled — they received the token. This is the thing we’ve been building toward: waiting should pay you, and you still get your tokens or sell them at the price you wanted. I had this crazy idea when I was searching for the project name — “to fave” means you earn while you wait to buy or sell. Like “I faved 85 SOL until the end of October for $153 payout” or “I faved $5,000 to buy 5 ZCash at the end of October at $1k and received $148 upfront.” MCP is 90-ish % done. MCP server = done, landing page = done, both not live yet. Last piece — a skill so agents can install and use the Faven MCP. Last 20% of any initiative is the hardest line to cross. On the video that got banned in under 10 minutes on LinkedIn: we killed the idea. Support confirmed we breached some of their rules. WTF: This week we had a bug in the web UI — we showed a higher premium than the user actually received in their wallet. I had already onboarded one user with this bug. He saw $1.01 and received $0.26. I fixed the bug and messaged him: “man, we lied to you about the premium, I just transferred you 0.75 USDC to compensate the difference.” We’ve got 4 waitlist signups in the last 24 hours. I’ve sent multiple manually prepared responses to see which one works best. The sample for A/B tests is basically nonexistent, but testing a few different messages and CTAs is at least some diversification. Want to join the party? DM or reply, something like "want to join".