M R P O N G

@SamPartee @openclaw @TryArcade I think this has to be the way. I’ve got a Mac Studio on the way

@PongInit @openclaw @TryArcade mac pro dedicated.

@openclaw ... but give it everything! Wrote a plugin to give OpenClaw / Moltbot / Clawdbot access to all of @TryArcade tools. Now you can use 100s of services from any of your msg apps i.e. Discord, Telegram, etc. Auth included 👇Me using GCal and Slack from Telegram

Someone please explain what’s awesome about MCP Apps - it feels like 2026 iFrame. What’s the point of generating the artifact if Claude can no longer interact with or manipulate the results? Consider: 1. I ask Claude to make a Figjam, awesome 2. Figjam needs tweaking 3. I either go into Figjam and lose Claude’s intelligence, or 4. I tell Claude to make another Figjam, and now I’m creating file hell Rinse and repeat for Amplitude charts, and other tools with dynamic assets. I mean I can see some value with just retrieving context for viewing like Box, but aren’t we just in Glean territory then? Why render that as HTML and risk data corruption or rendering problems? MCP Apps seem like a solution looking for a problem - the whole point of MCP is to have the actual data and actions at your agent’s fingertips so it can create (and iterate on) the asset better than downstream tools ever could

@ryancarson So if the machine writes your code, and your specs, and your tests - whis actually the creator? Might as well use Lovable if you’re going to reduce your contribution to a prompt

If you can do this, you're in the top 1% of engineers right now. Most engineers in enterprise are barely using agents (and if they are, most are stuck with copilot). If you can add looping at night, you go next level. It's not hard though. Just point your agent at this article (or copy/paste) and say "Help me set this up". Trigger the crons manually and iron out any bugs, then set it and wake up tomorrow to see what you've got.

One more thought for tonight: I’m coming to the realization that we are on the cusp of collapsing all of the “work about the work”. Sprint planning rituals, slide deck roadmaps, weeks of scoping/estimation, etc just feel crushingly slow/inefficient because of LLMs. The future of building is simply a clear plan - everything else is work that agents can do better than humans can.

Jan 24 update: Turns out that a simple Google Drive read permission could steal my whole day prepping for CASA Tier 2. Learned how to implement SQL-Cipher, hardened a bunch of security things and already passed my scan and self assessment. Gemini 3 Flash was the star today

@tgeisenheimer @aikarjal Hey Tim, I did it for you - it’s built on rust and MCP and it’s free forever. Enjoy at getminna.app

@aikarjal Been thinking about this exact problem and building an internal tool around it. In order to productize it you need to build pretty robust integration coverage. I would buy vs build if I could :).

Someone needs to build a company around Customer Context Graph. Collect all the threads – emails, meeting transcripts, slack messages, contracts, deliverables, detail, info, and config – from your customers into context that can be explored and queried by agents. This info is scattered between CRMs, ticketing systems, note takers, product, landing pages – it's inherently cross platform information. You need a new solution. Kind of how Segment did it trad SaaS apps. With this context, you can fire up Claude Cowork or similar for ad-hoc work or build extremely powerful agent automation flows. Expose the context as skills, MCP, and file system. Even better if you build it as open-source with a hosted option so people can take it on-prem as needed. Create a connector ecosystem around it. This will power every single next-gen AI-native full-stack business. Sort of like the context graph (@ashugarg @JayaGup10 ) that has been discussed recently but I'm thinking something very concrete: "Get me all the context about this particular customer." A customer-level, cross-system context substrate that agents can explore and act on

Yes, Zapier has the connectors but that’s just one part of the story. Problem 1: Zapier is built for APIs - JIT fetch of information with no graph, no history. Problem 2: Zapier is built for the org, not user - bot tokens galore - so it’s really expensive and complicated to build a personalized graph RAG in the cloud (lots of RBAC/sync issues) Problem 3: Cloud latency. Even if they did all that, context retrieval would be dog slow (multiple seconds) and grind agent workflows to a halt (not to mention a security nightmare). Problem 4 (biggest one of all): incentive. Agents kill all of the automation logic of a good IaaS like Zapier. If they build this, all of the work they have done to expose no-code actions to the user are useless. But what if they just build an agent platform to replace? Okay, but between the context graph and agent building you’re basically creating a brand new company that doesn’t leverage Zapier’s existing strengths at all

isn’t this just connectors → normalizer → graph db with summaries agents can query or pointer to info they can drill down grind would be connector maintenance which u *could* have agents just verify on a cron maybe but zapier could just launch a “customer graph” step and kill the startup unless u got vendor lock in by having a better agent layer if its open source mcp for Claude code, maintainer burden is probably too high If its paid mcp for Claude code, id predict a short life span

@aikarjal Please try getminna.app - it’s what glean should have been for agents and it’s OSS MIT license, free forever

Realization: the “AI copilot” era (2022-2025) was built for 80/20 human involvement, because early models just weren’t good enough to handle the bulk of the work. 2026 is the “Human Supervisor”era - the best builders are steering 80/20 agent first work and getting 4x return

Building a local context engine with Rust was cool. Then I realized: don’t be a dum ingestion pipe, make a graph. But then: graphs are dumb if they don’t have relevance. So I did that too. With Claude Code on my phone. While eating ramen with my wife github.com/getminna/minna…

Jan 20 update: -it’s alive! GitHub.com/getminna/minna… or brew install getminna/tap/minna-core -spent a ton of time today on tweaking TUI and dreaming up a better way to do graph-based priority for delta syncs. I call it gravity well and of course it uses context in the algo

can you guess when I first started working on getminna.app

jan 19 update: -realized that i'm overindexing on end user apps. if i'm going to build and validate this idea in public as OSS, the best place to start is a CLI -refactoring my OSS project today and getting ready to unveil tomorrow for many smart people to critique (i hope)

Jan 18 update: -learned way too much about public oauth apps and CASA and public declarations of privacy -redesigned FTUX to deliver value in 5 mins or less -cleaned 8ish GB of legacy code (man the early stuff was bad) -burned through every AG model for the week (god bless CC)

jan 16 update: -made a claude skill to generate "source" cover art with nano banana pro following my brand guidelines -planned out how to make an integrations factory to cover the top 50 sources that matter -finally got my rust backend and swift UI to be friends

if you suffer from the same problem, getminna.app

all of it to solve one major question: why do my AI agents have amnesia? every new conversation is a reboot and "JIT RAG" by uploading files directly or through integrations are spotty there should be a super fast, super simple way to suck in all the data from your work apps

Since Christmas: -Generated 50k lines of code and dove deep into Rust, Swift, Typescript -Create a MacOS app, Vercel auth bridge, Typescript/React marketing page, and in progress on an integrations catalog and iOS app -Wrote 103 docs with about 10 versions each