Skip to content

Quickstart

  1. Create a workspace. Sign up. Your first sign-in creates a workspace with a project from the Agent kanban template: a Features swimlane and a Bugs swimlane with sensible flow rules, plus a “Say hello from your agent” card to test with. Your first workspace also gets 3 days of Pro, no card needed. There’s no template to pick now; you pick one when you create another project. Joining from an invitation instead? You land in that workspace and on the same connect screen, so you can connect your own agent.

  2. Get your connector link. The connect screen asks which app you use and shows only that app’s steps. Every client connects through your workspace token’s private link:

    https://api.yokka.ai/mcp/c/<token>

    The link contains the token, so treat it like a password.

  3. Add the link to your agent. The connection is called yokka. If you belong to several workspaces, each gets its own, named yokka-<workspace> (Yokka <workspace name> in the chat apps), so they don’t collide. The connect screen shows the name to use.

    Terminal window
    claude mcp add --transport http --scope user yokka <link>

    If you added a yokka server before, the command fails because the name is taken. Remove the old one first, then add it again:

    Terminal window
    claude mcp remove --scope user yokka

    Optionally, have Claude Code check the board before each turn ends, so it never stops with a card left in progress and nothing said about it. Add this Stop hook to ~/.claude/settings.json (use your connection’s name as server), or copy the prompt from the connect screen and Claude Code adds it for you:

    {
    "hooks": {
    "Stop": [
    {
    "hooks": [
    {
    "type": "mcp_tool",
    "server": "yokka",
    "tool": "stop_check",
    "input": { "stop_hook_active": "${stop_hook_active}" },
    "timeout": 30
    }
    ]
    }
    ]
    }
    }

    See before you end your turn for what it checks.

  4. Say hello. Ask your agent to pick up the next card on the board. It should claim Say hello from your agent, report progress once and complete it. The connect screen lights up when the first MCP call lands, and the card moves to Shipped on its own.

  5. Make it stick in your repo (optional). Every agent already gets Yokka’s instructions when it connects: track the work you ask for as cards, claim before starting, wrap up before stopping. What the server can’t know is which project a repo belongs to. Once connected, the connect screen offers Copy setup prompt. Paste it into your agent inside a code repo and it adds a short section to AGENTS.md (or CLAUDE.md) that names the project. The section never contains your token, so it’s safe to commit. It’s also under Project settings → Prompts → Repo instructions, where you can copy the section to paste by hand:

    ## Task tracking: Yokka
    Work in this repo is tracked on the Yokka board, project "My app" (`my-app`), through the `yokka` MCP
    server. Pass `my-app` to tools that take a project.
    - Before starting non-trivial work, find its card (find_cards, get_next_card), or file it with add_card, and claim it before you start.
    - For everything else, follow the Yokka server's instructions.
    - If the Yokka tools aren't available, tell me instead of skipping the board.

    With more than one project, you can also connect the repo with a token limited to its project, so its agents can’t reach the others at all. See Project-limited connections.

Not ready to connect an agent? Watch a demo agent on your board, on the connect screen, runs a built-in agent on your hello card, so you can see a card move without setting anything up. Or choose Skip, show me my board: you can connect an agent any time from the board.

  • Learn how flow rules decide where cards go.
  • Read the agent loop to see the calls a well-behaved agent makes.
  • Connect a runner to start Claude Code or Codex on your machine from a card.