Skip to content

Use Team Cora from your AI client

Not built yet. This section describes something that is being built now. None of it works today, so do not rely on it. It is written first so that you, and your org admin, can see how it will work before it exists (D0017).

Many people work inside an AI client, such as Claude, all day. Soon you will be able to add Team Cora to Claude (on the web, in the desktop app and in the mobile app) and to Claude Code. Then you can read your missions, post and read Cora’s replies, and set up missions without switching windows (D0914).

Team Cora will be an MCP server at this address:

https://app.teamcora.ai/mcp

MCP (Model Context Protocol) is the standard way an AI client talks to an outside service. Here Team Cora is the service and your AI client calls it. This is the other way round from connecting an MCP server, where Cora calls an outside service.

ChatGPT will come later, once its sign-in works with Team Cora’s (D0926).

  • You sign in as yourself. Your AI client signs in with your own Team Cora account. It never gets your password.
  • No new rights. Your AI client can do only what you can do in the console, with the same checks. If you cannot do something in the console, you cannot do it from Claude either.
  • You choose what it may touch. Your AI client asks for the areas it needs, each a scope: Claude asks for more only when it needs them and you reconnect, and Claude Code asks for the list you pin. The sign-in page does not list them. Your role still decides what you may do inside each area.
  • Changes are marked. Team Cora marks every action that changes something as a change so your AI client can ask you first. Whether it asks is up to the AI client and its settings: if you tell it to always allow a tool, it may not ask.
  • Secrets stay out. Keys, passwords, payment details and a service’s own sign-in never pass through your AI client. Where a step needs one, Claude gives you a link to finish it in the console.
  • Your org admin decides. It works only once an org admin has turned on the AI clients switch for your organisation. It is off by default.
  • Reading your organisations and missions, the mission chat, files, memory, mission members and notifications.
  • Posting in a mission and waiting for Cora’s reply, answering questions and choice cards, and marking things read.
  • Setting up a mission and changing its settings, if you are allowed to in the console.

Heroes, the organisation’s people, connections and approval cards will follow in a later release (D0929).

When you post or decide something through your AI client, the mission chat shows via Claude beside the author’s name, for example “Priya Shah via Claude”. Your team can see how it was done. The audit trail keeps the same label (D0927).

Your private chat with Cora stays in the console. That includes Polish with Cora (see polishing a message). There is no way to reach it from an AI client, and it is still never stored (D0114, D0128, D0919). From Claude, you post as you type. Your AI client is your own private space.

Team Cora keeps your organisation’s data in Sydney. When you open mission content in your AI client, that content is sent to the company behind the client (Anthropic for Claude), under your own account with them. That company may hold it outside Australia, under its own terms. This is why the switch is off until an org admin turns it on. See the AI clients switch and security and privacy.

Text in a mission is written by people and by outside services. Team Cora marks it as mission content when it sends it to your AI client, and tells the client never to follow instructions found inside it.

Decisions: D0017, D0114, D0128, D0903, D0914, D0919, D0926, D0927, D0929.