From an ‘LLM wiki’ to team knowledge: scaling the context your agent reads

TL;DR — Lots of people now hand-build an ‘LLM wiki’: a folder of markdown files that a coding agent like Claude Code reads. It's powerful, but it usually stops at one person. Here's how to scale the context you gathered by hand to your whole team.

What is an ‘LLM wiki’?

A pattern has spread among developers lately: write your product, decisions, and rules down as a few markdown files, keep them in one folder, and let Claude Code or a local agent read that folder. People call it an ‘LLM wiki’ or a ‘second brain’ for their agent.

The idea is simple. Instead of explaining the background to your agent every time, you write the background down and say “use this.” From then on, the agent answers from the context you wrote, not from a guess.

Why it works so well

The point is that it gives the agent the right context. It isn't that the model got smarter — it's that you hand-picked the context, so the answers fit your situation. Organizing it by hand is itself an act of choosing what matters, and that shows in the quality.

That's exactly why a hand-built wiki is powerful: it isn't scraped clutter, it's context a person judged and chose. That's a strength to keep, not something to replace.

The catch: it stops at one person

But a hand-built wiki usually lives on one person's laptop. A few things follow from that:

• Teammates can't read that context, so everyone ends up building their own wiki.
• It's only as fresh as one person keeps it. When they're busy, it goes stale.
• Someone else's agent — or an agent on another machine — can't reach that folder.

What works great for an individual hits a wall the moment it needs to work for a team. Context is only powerful when it's gathered in one place; scattered across personal folders, you're back to the original problem.

How the team version is different

Specify takes this ‘LLM wiki’ idea as-is and grows it to team scale. The idea is the same: gather the context your agent reads into one place. The difference is that the context lives in your team's shared workspace, not one person's folder.

• It gathers itself. Beyond docs and decisions, connect Notion, Slack, and GitHub and their context comes along too — no copying by hand.
• You share it. The context one person organized is read by every teammate and their agents at once.
• It connects straight to your agent. Specify runs as an MCP server, so Claude Code and Codex explore your team's context directly — no copying a folder around.

It's the same work you did by hand, moved to team scale.

You don't have to throw away what you built

If you already have a wiki, that's a great starting point — no reason to toss it. Bring your existing docs and tools in by connecting them, and just fill the gaps. You don't have to move everything at once.

Start small. Gather the one thing your team keeps re-asking into context, and connect your agent to it. The feeling you got from a personal wiki — not having to re-explain every time — arrives soon, this time for the whole team.

Frequently asked questions

Can I bring my existing docs and tools?

Import your documents into the workspace, or connect the tools you already use as sources, and their context comes along. You don't have to rewrite anything from scratch.

Can I keep using it from Claude Code?

Yes. Connect Specify as an MCP server in one line, and Claude Code reads your team's docs, decisions, and code context directly. Your workflow stays the same; only the context becomes your team's.