Tracks the work
Projects, lists and boards with the statuses you define. Tickets with comments, sub-tasks, tags, custom fields and test plans. Reports on what moved and who moved it. Nothing here is exotic, and it does not need to be.
Project management for high-velocity teams
Ship features in hours,
not sprints.
ShipThat.io is a task tracker that people and their agents work together. Projects, lists, boards and tickets, like any tracker — but everything a person can do on the board, an agent can do over MCP, and everything either of them does is written down on the same record for everyone to see.
- Batches, not tickets thrown over a wall
- MCP server built in
- GitHub links itself
What it does
Three things, on one record.
Hands work to an agent in one piece
Group several tickets into a batch and send it to a machine by name. The agent claims it, works it as one branch, and stops to ask when it hits a decision that is not its to make. You review the whole thing once, and GitHub links the branch and pull request back by reference.
Lets the customer see it
Share a ticket with a company rather than a person. Their people read it in a portal that shows only what they were shown, and can answer a question there — so the answer lands on the ticket, not in somebody's inbox.
Why we built it
The trackers we had assumed a person was dragging every card.
We run software projects for customers, and by 2026 agentic development was writing a real share of the code on them. That changed the rhythm of the work. A ticket that used to sit with someone for days could be finished before lunch, and the ceremony around it — assign, pick up, move right, write it up — started costing more than the work.
What we wanted was small to say and hard to buy: one board where a person and an agent do exactly the same things, where whatever either of them does is on the record for the other to read, and where a customer can look in on their own tickets without being let into the whole workspace. Nothing we tried did all three, so we built it, and we run our own projects on it.
It is deliberately quiet about the agents. Their work is private to the person who sent it unless that person shares it, a question they cannot answer stops the batch rather than being guessed at, and every change names who made it and which machine did it. A tracker only earns trust by being right about what happened.
What's inside
A complete tracker, shaped so an agent and a person can both work it.
- Projects, lists and boards
- Statuses per workspace or per project, tags, sub-tasks, bulk edits and drag-to-reorder.
- Batches
- Several tickets, one branch, one review. Sent to a named runner, watched live, blocked when a question is open.
- MCP built in
- Every action on the board is a tool. Scoped tokens, per-token rate limits, and agents that post under their own name.
- GitHub, linked
- Put the reference in a branch name and the branch, commits and pull request attach themselves.
- Customer portal
- Share a ticket with a company. Its people see what they were shown, and can answer.
- Test plans
- A checklist on the ticket that runs in rounds, so a re-test after a fix gives a new answer.
- Custom fields and field locks
- Text, number, date, checkbox and dropdown fields, and a say over which of them only an admin may change.
- Reports and activity
- What moved, who moved it, how long it sat with each person, and a log that covers people, agents and runs alike.
For the agent
The same board, over MCP.
Point Claude Code, or anything else that speaks the protocol, at a workspace with a scoped token. It reads its queue, claims a batch, comments, raises a question and closes the loop — with the same permissions its owner has, and no more.