Tavimi

Getting started

From sign-up to your first connected AI agent, in ten minutes.

Three steps: create your account, connect your assistant, then, with Claude Code, install the plugin and its skills to run a space.

  1. Create your account and your first space

    An email address, a password, then a second factor: an authenticator app or a passkey. Then create your organization and one space per project.

    Sign up

    The Tavimi web app is in French for now; an English interface is on its way. Your assistants work in any language.

    Prefer some guidance? Request a pilot: we set up your spaces and your first tasks with you.

  2. Connect your assistant

    On the first call, your browser opens: you sign in, then choose the spaces, or entire organizations, the assistant will be able to see, and whether the connection is read-only.

    Claude on claude.ai and Claude Desktop, ChatGPT, Le Chat: add a connector with this address

    https://app.tavimi.com/mcp

    On claude.ai and Claude Desktop, Tavimi is also listed in Claude's connectors directory: add it from its listing, no address to copy.

    Depending on your plan, ChatGPT first asks you to turn on developer mode; in Le Chat, an admin of your account adds the connector.

    Codex, by OpenAI

    codex mcp add tavimi --url https://app.tavimi.com/mcpcodex mcp login tavimi

    Claude Code, in the project folder

    claude mcp add --scope project --transport http tavimi https://app.tavimi.com/mcp --header "X-Tavimi-Projet: my-project"

    Replace my-project with the folder name: that way each project keeps its own connection. Restart Claude Code, then type /mcp to authenticate.

  3. Install the Tavimi plugin

    In Claude Code, two commands: the first adds the source, the second installs the plugin. You need both.

    /plugin marketplace add software-family/tavimi-plugins/plugin install tavimi@tavimi

    The server must be named tavimi, as in the previous step: the skills refer to it by name.

What the plugin adds

Skills to run a space

  • Setup

    An interview to organize Tavimi for your team: spaces, task statuses, roles, note categories. The assistant proposes a plan, then creates what you approved.

  • Questions

    The assistant reviews open tasks and asks, on each one, the questions that clear up its ambiguities, before anyone starts on it.

  • Next task and loop

    The assistant picks a ready task and takes it all the way to a pull request, or asks the missing questions. The loop works through tasks one at a time; nothing is merged without you.

FAQ

What people ask us

Why an X-Tavimi-Projet header?

Claude Code keeps a single connection per server configuration: two projects declared the same way share one connection, and the latest consent applies to both. A different value per project, such as the folder name, gives each one its own connection and its own spaces. Tavimi does not read this header.

Can I change the spaces of a connection?

Not after the fact: revoke the connection from the “Mes connexions” (My connections) page, then sign in again from Claude Code with /mcp. The consent screen asks for the spaces again.

claude.ai keeps only one connector per address: how do I separate two uses?

With connection profiles. The owners of your organization create them: each has its own MCP address, its spaces and, for each space, read or write access. They assign them to members, who find the addresses of theirs in “Mon compte” (My account) and in the Agents tab. One connector per profile, and each use keeps its own scope.

Is the plugin required?

No. Without it, your assistant reads and writes in Tavimi just the same. The plugin adds skills to run a space: setup interview, questions, next task and loop.

What about assistants other than Claude?

ChatGPT, Codex and Mistral's Le Chat connect to the same address, like any MCP client that supports OAuth: the Agents tab of your organization gives step-by-step instructions for each. The plugin itself is specific to Claude Code.

Ready to start?

Create your account, connect your assistant: in the next conversation, it picks up where it left off.