AI Agents (MCP)
Let your coding agent keep Kestrio tasks, sprints and time up to date — with your approval for every change.
AI Agents (MCP)
Connect the AI agent you already work with (Claude Code, Cursor, or any MCP client) to Kestrio. While it works, the agent can find the task the work belongs to, propose new tasks in the right project and sprint, update status with the merge request link, log your time when you ask, and remind you when a sprint is ending.
Nothing changes silently. Every change is shown to you first, and the agent makes it only after you say yes.
Connect an agent
-
Open Settings → AI Agents and copy the setup command, for example:
claude mcp add --transport http kestrio https://<your-kestrio-domain>/api/mcpCursor and other MCP clients use the same URL.
-
The first time the agent uses Kestrio, your browser opens. Sign in, pick the workspace (organization) the agent should work in, and choose Allow. If the workspace requires two-factor authentication, you complete it first.
-
To stop an agent, select Revoke next to it in Settings → AI Agents. Access ends immediately.
Each connection is bound to one workspace. To use another workspace, connect again and pick it.
What the agent can do
| Ask for | What happens |
|---|---|
| "What am I working on?" | Lists your projects, active sprints, open tasks and notices. |
| Starting a piece of work | Searches for a related task first; if none fits, proposes a new one. |
| Creating a task | Goes into the project's active sprint. It goes to the backlog only if you say the work is planned for later. If there's no active sprint, the agent asks you. |
| "Start this task" | Moves it to in progress and assigns it to you. |
| "Update the task" | Changes status and adds a note, e.g. the merge request link. |
| "Log my time" | Adds a draft entry to your timesheet. Without a duration, the agent suggests the time since you started the task, and you confirm or correct it. You submit the timesheet yourself, as usual. |
| Sprint ending soon | The agent mentions it once and can plan the next sprint, with the same projects, length and board columns. |
Permissions
The agent can never do more than you can in Kestrio:
- If you can only propose tasks, the agent's tasks go to the approval queue as drafts.
- Logging time needs timesheet access; planning sprints needs sprint management.
- On the consent screen you can see exactly what the connection may do: read, tasks, time, sprints.
Every change made through an agent appears in the audit log as made by you, through your agent.
Tell your agent how your team works (optional)
Agents also follow instructions in a repository's AGENTS.md or CLAUDE.md. A good default:
## Kestrio (task & time tracking)
- At the start of a work session, call `get_my_context` with this repo's git remote URL.
- Before starting work, call `find_related_work`. If a task matches, confirm it with me; otherwise propose a new task and ask before creating it.
- One task per user-visible piece of work; use subtasks for steps.
- New tasks go into the active sprint; use the backlog only when I say the work is for later.
- Never log time unless I ask.
- Show me every preview and call `confirm_action` only after I say yes.The first time you work in a repository, the agent may ask which project it belongs to. After you confirm, it remembers.