How I run my life on Markdown and the Stacklist MCP

Everyone I know is drowning in inputs.

Slack threads. Gmail. Calendar invites. Granola meeting notes. Linear issues. Lightfield CRM follow-ups. Text messages. WhatsApp. Each of them authoritative about a slice of your life. None of them talking to each other. All of them expecting you to be the integration layer.

I'm a founder and a product manager. I don't want to spend my morning bouncing between tabs to figure out what today actually is. I need two things, and I need them done reliably:

1. Absorb the context from every one of those sources. Nothing important slips through.
2. Distill it. Every project, every initiative, every loose commitment in one place, shaped into the handful of things that actually need to be done today.

The orchestrator is a Claude Code project I call Virgil. The inputs are the eight apps above, reached through MCP servers. Everything lands in Stacklist — our startup. Stacklist is a discovery network for AI that also serves as the central repository of knowledge for an individual person. I happen to run my entire operating system on top of it.

The closest pop-culture shorthand for what this feels like is wake up, Jarvis. Minus the suit. Plus a lot less drama.

This piece is the wiring diagram.

The wiring

Virgil is the orchestrator. It reads from eight sources through their MCP servers, distills what it finds, and writes the result into a single hub. Reads on one side, writes on the other, distillation in the middle.

Stacklist is where everything lives after that. One place to point my attention. One place to come back to. One place where the work of absorbing and distilling has already been done.

The rule

One line, verbatim from the CLAUDE.md Virgil loads at startup:

All working state goes to Stacklist via MCP. Local files are temporary only. Every Markdown file created goes into Stacklist, organized by week.

One consequence matters more than the others. When everything for a week is already in Stacklist — daily cards, meeting summaries, decision docs, brain dumps — I can ask anything to synthesize it. Virgil from Claude Code. Claude on desktop. Stacky, the AI built into Stacklist itself. All three are reading the same hub. All three give me consistent, current answers, because there's only one place the week's work lives.

Most AI-plus-notes setups bolt an LLM onto a tool built for humans. The AI is a visitor. Here the AI is a first-class API client, using the same MCP endpoints any integration would. The human-facing UI on stacklist.com is a sibling, not a parent.

Why Stacklist and not Notion, Obsidian, or a folder of Markdown

Every developer reading this has the same question. Here's the short version.

Notion. The AI is a sidebar. You prompt it, it answers, the session ends. It can't wake up at 7 AM, read your eight sources, and write a daily card while you sleep. There's no API surface for an outside agent to act as a first-class user of your workspace.

Obsidian. File sync between devices. That's the product. No shared hub for multiple AI clients to read and write against. No semantic search that a CLI tool can call. No write-API for an agent to use while you're offline.

A folder of Markdown. Works fine for you alone. Breaks the moment you want Virgil (Claude Code), Claude desktop, and Stacky (the AI inside Stacklist) all reading and writing the same place with consistent state. Local files can't be that place. One device, one process, one direction.

Stacklist. A hub with an MCP server. Cards and stacks are first-class objects over HTTPS. Any client with a key — an agent, a desktop app, the web UI — reads and writes through the same endpoints. Semantic search and keyword search ship with the product. The AI doesn't visit the hub. It uses the hub the same way I do.

The point isn't that Notion and Obsidian are bad. They're good at what they're for. What I needed was a hub that treats an AI agent as a peer user, not a plugin. That's a platform problem, not a notes problem.

The daily routine

Every morning starts the same way.

At 7 AM ET, Virgil wakes up and runs its standup routine. It reads yesterday's daily card, then walks the eight sources one by one — meetings from Granola, overnight email from Gmail, today's calendar, Slack mentions, Linear moves, Lightfield follow-ups, texts and WhatsApp.

Each source gets a read. Then Virgil writes one daily card. One Markdown file in Stacklist. Today's date as the title.

By the time I open my laptop with coffee, the card is already there. A focus list at the top. An overnight-inputs digest in the middle. A spoken wrap-up at the bottom that tells me which item to do first, which to do after lunch, and which to batch until Friday.

I didn't open Slack. I didn't open Gmail. I didn't open Granola. Virgil opened them for me. The parts that mattered are on the card. The parts that didn't stayed where they were.

The handoff trick

The mechanism that makes it work across days is small and almost embarrassing to describe.

Every daily card ends with three subsections under ## Handoff Out:

• Completed — what got done
• Still Open — what didn't
• Tomorrow — what's planned

The next morning, Virgil generates the new daily card by pulling yesterday's Handoff Out into today's Handoff In. Monday reads Friday. A missed day walks back through the week's stack to the most recent card. A longer gap falls through to the most recent weekly synthesis.

The AI never loads the whole week. Just yesterday's tail, plus today's fresh inputs.

Every day ends with a Handoff Out. Every day starts with a Handoff In. I don't re-explain anything at 7 AM. Yesterday's Kyle already did.

It's the cheapest context-window trick I've used. It also happens to be what writing things down has always been for.

Honest pitfalls

I'd rather name the rough edges than pretend they don't exist.

The auto-summary can hallucinate. Stacklist runs an LLM over the card body to populate a preview summary. It once confidently described a daily card as being about "Kyle Campbell." My name is Kyle Hudson. Fix: anchor identity in the body, don't rely on inference.

Brain-dump discipline is the bottleneck. There's a Granola folder called "For Virgil" where I dictate a 2–5 minute brain dump most mornings. Skip it for two days and the system has to nag me back. No AI guide works without input.

The Todoist pile still grows. A typical standup surfaces 20 to 30 overdue items. The system names the pattern and recommends triaging, delegating, or batching. The discipline is still mine.

See what a weekly OS stack actually looks like

I put together a companion stack on Stacklist that shows the structure: a README card, four example daily cards (Monday through Thursday), Handoff In / Focus / Working Notes / Handoff Out sections, pattern flags, all of it. Names and numbers are scrubbed. The shape is real.

Virgil Weekly Stack — Example Week: stacklist.app/c/technology/stack/f92fd9d2-aed7-49cc-90ff-2f4288dd2a28

If the diagram above is the wiring, the example stack is the house with the lights on. Poke around, read a daily card end to end, see how the handoff chain reads when it's running.

The trick

The trick isn't the wiring. It's what the wiring does to your day.

Eight inputs go in. Virgil reads them. One hub holds the result. One daily card distills it. Three priorities come out.

Markdown is the format. MCP is how the AI reads and writes. Stacklist is where it all lives.

When the AI shares your hub instead of visiting it, the hub gets to be the operating system. That's the part you can feel the first morning you open your laptop and the work has already done itself.