Guide
What is TwinMCP, and why?
After this page you'll know what we do, where you fit and what to do next. A 5 minute read.
The problem
An expert's most valuable knowledge isn't in a textbook. It's the small calls built up over years: “If a client wants a fixed price, I propose paid discovery first, because…”, “With this kind of car I check this part, not the inspection report…”. That knowledge lives in their head and doesn't scale. They can only talk to one person at a time.
AI assistants talk about everything, but generically. They don't know what someone with 15 years in your field would do in this exact situation, and they don't tell you when they don't know.
The answer
On TwinMCP experts teach how they decide to a digital twin. The twin works from cards made of the expert's own decisions and plugs into the AI agents people already use. Users don't learn a new app; when they ask Claude or ChatGPT a question, the agent consults the right expert's twin when needed.
For experts: teach once, take part in hundreds of decisions a day. For users: a real expert's stance instead of a generic answer.
How it works
- 1Expert
Teaches their decisions
Turns the calls they make in interviews and daily work into decision cards: what they do, why, and when they wouldn't.
- 2Digital twin
Answers from the cards
Answers questions from the expert's cards with a stance, reasons and a confidence score. It says so when no card covers the topic.
- 3The user's agent
Consults when needed
When the user asks a work question, Claude, ChatGPT or Cursor consults the right twin over MCP and brings the answer into the chat.
- 4The loop
Demand grows the twin
Topics where the twin is weak are clustered anonymously and come back to the expert as training tasks. Each answer sharpens the twin.
Words you'll see
Terms that show up in the app. No need to memorise; come back when you get stuck.
- Digital twin
- An AI that carries an expert's way of deciding. It is not the expert; it speaks from their cards.
- Decision card
- The unit of a twin's knowledge: a situation, the expert's stance, the reason and the exceptions. More and sharper cards mean better answers.
- Interview
- The starting step: 10 dilemmas from the expert's field. Each answer becomes a card.
- MCP
- Model Context Protocol, the standard way AI agents connect to outside tools. You connect TwinMCP with one URL and twins show up as tools in your agent.
- Agent
- An AI assistant that can use tools, such as Claude, ChatGPT or Cursor.
- Library
- The experts a user attaches to their agent. Up to 8 twins appear as separate tools.
- Coverage label
- “Grounded” means the answer rests on the expert's cards; “extrapolated” means there's no card on the topic.
- Training task
- An anonymous summary of a topic at least 3 different people got stuck on. The expert answers and approves; the twin updates.
- Maturity
- A 0-100 score from the number of cards, how much of the field they cover and evaluation results.
- Credits
- Each answer costs a few credits. In beta everyone starts with 500 free credits; no real payments.
What it isn't
Not a professional service
A twin's answer is an opinion, not a contract, diagnosis or legal advice.
Not the expert
The twin makes no promises for the expert, takes no jobs and shares no contact details.
Not a chat character
It's for work decisions. Health, law and finance stay closed until legal sign-off.
Where this is going
Today we're in closed beta. The order below is a plan and may change based on what beta tells us.
- Now
Closed beta
6-8 experts, 50-100 invited users, 4 weeks. Credits are free, earnings simulated. We measure whether people actually use twins, whether answers sound like the expert, and whether experts answer tasks.
- Next
Real payments and payouts
Buying credits, real payouts to experts, more experts and categories. Experts pick their own rate tier.
- After that
Easier training
Answering training tasks by voice or messaging, team twins, company use.
Which one are you?
I'm an expert
I want to build my twin
What to do on day one, every week, and how to write a good card.
Expert guide →
I'm a user
I want to consult experts
Picking experts, connecting your agent, asking well and reading answers.
User guide →
Keep reading