MCP - Connect an AI Client

MCP (Model Context Protocol) is an open standard that lets AI assistants call tools in other applications. Decuga exposes an MCP server so any MCP-compatible client - Claude Desktop, Claude Code, Gemini CLI, or a custom integration you build yourself - can act on your behalf: creating and updating tasks, moving sprints along, triaging service desk tickets, and more, all described in plain language to whichever AI client you're already using.

This is the same set of actions available from the Decuga Slack assistant and the in-app chat panel - just reachable from outside Decuga too, through an AI client of your choice. Every action still goes through your own account and your own project permissions; nothing an MCP client does bypasses what you're normally allowed to do in the dashboard.

There are two ways to connect, depending on which client you use. Both end up in the same place - your AI client gets access to the same set of actions, scoped to your account.

claude mcp add decuga --transport http \
  --header "Authorization: Bearer dmcp_your_key_here" \
  https://decuga.com/api/mcp

If your team tracks work in Decuga, this optional skill has Claude Code check the current sprint for a task that already covers whatever it's about to commit - creating one if there isn't one - and put the task's code at the start of the commit message, like DT-142: Add retry to the webhook sender. The code goes first because git tools and the GitHub UI cut long messages short, and the start is the part that always shows. Every commit ends up traceable to a real task, without you having to look one up and paste the code in yourself.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "if": "Bash(git commit*)",
            "command": "jq -n --arg msg 'Before this commit lands, apply the decuga-task skill if a Decuga MCP connection is available in this session.' '{hookSpecificOutput: {hookEventName: \"PreToolUse\", additionalContext: $msg}}'"
          }
        ]
      }
    ]
  }
}

Every personal MCP key you create is listed on your Profile page, along with when it was created and last used. You can create as many as you like - one per client or device is a good default, so you can tell them apart and revoke one without affecting the others.

A sample of what you can ask - not an exhaustive list. Type naturally; these are just starting points.

Unlike the in-app chat panel - which is always scoped to whichever project you have open - an MCP client isn't tied to one project. If you belong to more than one, just ask it to list your projects, or name the one you mean and it will resolve it for you.

Every action still checks your actual membership and role on that specific project, the same as if you'd clicked through to it in the dashboard yourself.

An MCP client acts as you- it has exactly the permissions your account has, nothing more. Every tool call goes through the same authorization checks as the dashboard and the Slack assistant; there's no separate, more-permissive path for MCP clients.

• Personal API keys can be revoked any time from your Profile page.
• Claude Desktop's connector flow uses your normal Decuga sign-in - Decuga never sees or stores your Claude credentials, and Claude never sees your Decuga password.
• Nothing an MCP client does is retried or applied silently on your behalf without you asking for it - the assistant only acts on what you tell it in that conversation.