Mindset AI

Docs / Getting oriented

What we bring in

Nobody starts from nothing. The material already exists, scattered across laptops, chat histories and half-finished projects.

What comes in, and what it becomes
What your people already builtWhat it becomes in MindsetA Claude or ChatGPT skillAn instruction file somebody tuned over monthsA Mindset agent your organisation owns, rather thansomething living in one person's account.An MCP serverOften on a laptop, often pointed at something realA Mindset connection, with each of its tools as a namedMindset operation.An API, database or spreadsheetHTTP, Postgres, Google SheetsA Mindset connection. Operations are discovered, andenabling one makes a real call to prove it works.Documents an agent readsA Copilot knowledge source, or a Claude projectA Mindset knowledge connection, so reaching it is recordedlike anything else.
Nothing is built from scratch. Four kinds of starting material, and what each becomes.

Skills and prompt files

A Claude skill, a long instruction file, a prompt somebody has tuned over months. This is the most common starting point and the easiest. The file gets pasted into Mindset, which reads the connections and operations that already exist in your Mindset organisation, proposes what to build, and waits until somebody approves the plan.

What comes out is a Mindset agent owned by your Mindset organisation, rather than something living in one person's Claude account. The instructions become the agent's standing instructions, the stages become an ordered list the agent works through, and anything the agent needs to reach becomes a connection with named operations on it.

MCP servers

Someone has an MCP server running, often on a laptop, often pointed at something real. The shape is the same. The server becomes a connection and each of its tools becomes an operation on that connection.

One thing a person does rather than Mindset. An MCP server publishes its tools without reliably saying which of them change the world, so somebody marks each one read or write. Err towards write.

  • A read marked as a write costs somebody an approval click.
  • A write marked as a read means an agent changed something in one of your systems and nobody was asked.

APIs, databases and spreadsheets

An HTTP API, a Postgres database, a Google Sheet.

  • Operations are discovered rather than hand-written.
  • Enabling one makes a real call to prove it works, and the output shape comes from that real response rather than from the documentation.
  • For a database, Mindset checks what the credentials you supplied are allowed to do before enabling anything, and proves the query works without changing any data.

Knowledge bases

Documents a Mindset agent needs to reason over. In Mindset a knowledge base is a connection like any other, so an agent reaching it does so through something named and recorded. Material that is currently a Copilot Studio knowledge source or a Claude project becomes a Mindset knowledge connection.

What does not come across: anything that works by driving a browser, and anything that has to resume after a failure. Both are boundaries rather than roadmap items.