MCP and Claude won’t solve your company’s knowledge problem.

In fact, they’ll probably make it worse.

The demo is a bit seductive. Connect a model to a few tools, point it at some docs, ask a question, and you’re done.

It feels like the stack is enough.

IMHO, that framing misses what teams actually struggle with. A company brain is not just access.

It’s context, trust, and structure. Without that, you don’t have shared knowledge. You have a smart interface sitting on top of chaos.

MCP + Claude can be a powerful layer. But the layer is not the system.

→ 𝗔𝗰𝗰𝗲𝘀𝘀 𝗼𝘃𝗲𝗿 𝗔𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁 If every source says something slightly different, the model will still answer. That does not mean the answer is grounded in a team-approved reality. A company brain needs one place where knowledge gets shaped, clarified, and maintained.

→ 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 𝗼𝘃𝗲𝗿 𝗥𝗲𝗹𝗶𝗮𝗯𝗶𝗹𝗶𝘁𝘆 Pulling information is easy. Knowing whether it is current, duplicated, or safe to act on is the hard part. Teams do not just need responses, they need confidence.

→ 𝗧𝗼𝗼𝗹𝗶𝗻𝗴 𝗼𝘃𝗲𝗿 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄 People assume knowledge work starts with a prompt. Usually it starts with onboarding, handoffs, decisions, and repeat questions. A company brain lives inside those moments, not just inside chat.

We should absolutely use MCP. We should absolutely use Claude. I love that direction.

But if we skip governance, curation, and usability, we are building a very fast way to surface messy information.

We’re currently writing an ebook on how teams build trust and structure in their knowledge bases.

I’m curious: is your team trying to assemble a company brain from model + connectors alone?

If you are, I’d would love to hear what is working, and where it breaks - it would help me a ton if you can fill in this 4-min survey or we’ll have a coffee chat about it

Share your experience here: https://docs.google.com/forms/d/e/1FAIpQLSfuhJuWQpqSw3QgRQg4XsSqsu0fuDcvf8b1f8-Yb6OzhQqL9A/viewform…

How does your team build its Company Brain?

Source: https://docs.google.com/forms/d/e/1FAIpQLSfuhJuWQpqSw3QgRQg4XsSqsu0fuDcvf8b1f8-Yb6OzhQqL9A/viewform Not at all (banned in our org)

Basic (I use ChatGPT/Claude on the web for basic drafting, emails, and brainstorming)

Pretty okay (Pro plan user) — I refine prompts, use custom instructions, and iterate multi-step queries

Pretty good (Claude Code/Codex user) — I have my own skill files and agents deployed

Native — I rely on interconnected AI agents that independently plan, execute, and self-correct tasks