PonosProduct updates
Run your backlog from Claude: Ponos's MCP now manages projects, goals and recurrences
Ponos's MCP connector can now create projects, goals and recurring tickets, tie tasks to a goal and launch them after a cost preview — all through the console's own rules, and never deleting anything.

A week ago, Ponos's MCP connector could read your board and write a draft ticket. This week it can set the board up. Here is what was merged on main since the last article, grouped by what it changes for you rather than by pull request.
If you have not met it: Ponos turns a ticket moved to Ready into a Claude Code session on a disposable git worktree, and brings it back as a pull request or a written page. The MCP connector is the door through which Claude (claude.ai, Claude Desktop or Claude Code) talks to your board.
Your assistant can now organise the board
Until now, creating a project, a goal or a recurrence meant opening the console. Four groups of tools change that, and all of them go through the console's own code path: same names to check, same validations, same errors. A client does not get a back door.
- Projects —
create_projectandupdate_project. Point to a git folder already on the machine, give a GitHubowner/namefor Ponos to clone, or create a document project with no repository. A taken name, a missing folder or a failed clone is refused and nothing is left half made. - Goals —
create_goalandupdate_goal: a name, a criterion, a due date, a priority from 0 to 100 and the projects it covers. A goal can be turned off; it cannot be deleted from here. - Tying tasks to a goal —
set_task_goalattaches existing tasks to a goal, or detaches them. Everything is checked before anything is written, so one unknown task refuses the whole batch, and the answer shows each task's goal before and after. - Recurrences —
list_schedules,create_scheduleandupdate_schedule.
You can now say, in a conversation: "Create a project for the repo acme/billing, a goal 'Ship invoicing by December', and tie these four tickets to it." Claude does it in a few calls, and you read the result on the board.
Recurrences never switch themselves on
A recurrence is the riskiest thing to hand to an assistant: it creates tickets by itself, on a clock, and each one can start a paid session. So the rule is blunt.
A recurrence written by a client is always born off. No tool turns it on, launches it or deletes it.
The console marks it To review and offers an Activate button. You read the mission, the cadence and the type, and you press the button yourself. That is the same posture as the rest of the connector: a ticket a client creates is a draft, and nothing deletes data.
Recurrences can serve a goal
Recurrences got a field of their own this week: Goal. A schedule that serves a goal ties every ticket it makes to it, whether the ticket is born at the appointed hour or from Run now. The goal's progress bar — n of m tickets done — then counts the recurring work too, instead of leaving it on the side.
If you run a weekly dependency bump or a weekly report, this is the missing link: the goal "Keep the product healthy" now reflects the chores that actually keep it healthy. A goal that is over, or deleted, does not stop the schedule; its tickets are simply born without one.
Launching, with a price tag
launch_task lets the assistant move tickets to Ready, which starts paid sessions. It works in two calls on purpose. Without confirm, nothing changes and you get a preview: the tasks kept, with project, model and estimated cost, and the ones refused, with the reason. Only after your explicit yes does the second call, with confirm: true, launch them. Every launch is written in the ticket's history under the client's name. If you would rather keep that gesture for yourself, switch off The assistant may launch tickets in the settings.
A ticket page that reads better
Two changes in the console you will see every day:
- The ticket page splits Mission and Result into two tabs. Mission is always there; Result only appears once the work is done, so the page of a running ticket no longer shows an empty slot.
- The On prompt, which Ponos runs at each change of status, was hidden under the journal. It now has its own field at creation and its own section in the Mission tab, and it can be written from the MCP as well.
Cleaner commits, and a disk that stays in bounds
Two quieter changes, both about trust:
- Sessions no longer sign commits and pull requests as Claude. No
Co-Authored-Byline, no "generated with" footer — unless you turn theclaude.attributionsetting on. Your history reads as yours. - Rotating logs. The
logs/folder never exceedsrunner.log_max_mb(4 GB by default, adjustable in the settings). Beyond that, the oldest session logs go first. A runner that lives on your machine all year should not slowly eat it.
Ponos also stopped committing agent-tool files to your repositories: instructions live in AGENTS.md, and a worktree that only has it gets a one-line, never-committed CLAUDE.md pointing there.
Try it
Connect Ponos to Claude once — Settings → Connectors → Add custom connector with https://<your console>/mcp, then Connect — and ask "what is waiting for me?". Then ask for a goal and a recurrence, and open the console to review what Claude drafted. Nothing it creates is switched on without you.
The installation guide takes a few minutes on Linux or macOS, and the MCP section of the console guide lists every tool and what it is allowed to do.
A Hermez product.