Agent connection

One URL, two jobs: consult experts, train your twin.

Add TwinMCP to your agent once. Experts in your library show up as tools. If you own a twin, the same connection also unlocks training tools.

What is MCP?

MCP (Model Context Protocol) is a common standard for AI agents to connect to outside tools. Once connected, your agent sees TwinMCP's tools and decides during the chat when to use them. You just talk normally.

Connect

Claude, ChatGPT, Cursor and VS Code ask you to sign in and approve; no key needed. For other clients, create an API key in My space → Connections.

  1. In claude.ai or Claude Desktop open Settings → Connectors → Add custom connector.
  2. Name: TwinMCP. URL:
    https://api.twinmcp.cronwork.com/mcp
  3. Click Connect, sign in with this account and approve.
  4. In a chat, make sure TwinMCP is enabled in the tools menu.

To consult

Everyone gets these. Your agent usually picks them itself; you can also ask by name.

ask_<uzman>
One tool per expert in your library (first 8). The agent calls it when the topic fits.
find_experts
Searches experts by topic.
ask_expert
Asks any published expert, even outside your library.
rate_answer
Rates an answer from 1 to 5.
  • ›A client wants a fixed price for an unclear scope. Consult the TwinMCP experts and tell me what to do.
  • ›Ask Doğu: no-code or custom build for a first MVP?
  • ›Is there a used-car expert on TwinMCP?
  • ›Rate the last answer 4.

To train your twintwin owners

Once you own a twin, studio_* tools appear on the same connection. The agent follows a training script: check status, ask, read your card back, save it if you confirm.

studio_status
Status and next step.
studio_daily
Today's training: a curriculum question, a demand task, an "Is this you?" check and your streak.
studio_interview_next
Next interview or advanced curriculum question (30 per field).
studio_capture
Turns a decision you mention during other work into a candidate card, if you say yes.
studio_teach
Saves a decision as a card. Published directly if you confirmed in chat, otherwise queued for review.
studio_proposals / studio_review
Lists, approves or rejects cards waiting for review.
studio_tasks / studio_answer_task
Fetches and answers tasks from demand.
studio_check_answer
Scores one of your twin's real answers 1-5; your correction becomes a card proposal.
studio_test
Asks your twin a free test question.
studio_publish
Publishes or cuts a new version (5+ cards).
studio_insights
Anonymous demand themes.
  • ›Interview me with TwinMCP. One question at a time; turn each answer into a card and get my OK.
  • ›Here's a call I made with a client today: … Teach it to my twin.
  • ›Fetch my training tasks and let's answer them together.
  • ›Ask my twin this and judge the answer against how I'd say it: …
  • ›What's my twin's status, is it ready to publish?

Troubleshooting

No tools showing
Check TwinMCP is enabled in your client: the tools menu in Claude, Settings → MCP in Cursor. Removing and re-adding usually fixes it.
Expert not a tool
Add them to your library; the first 8 become tools. Others are reachable through ask_expert. Some clients refresh tools only on reconnect.
No studio_* tools
They only appear for accounts that own a twin. Create your twin on the web first, then reconnect.
Cap reached
The connection's monthly cap is used up. Remove and re-add TwinMCP in your client and pick a new cap on the consent screen.
Your agent already knows how to use TwinMCP: the server sends it short instructions on connect. No need to memorise commands.

Keep reading