CEO
Opened PL‑7 — Scraper picks a junk first result straight out of the WhatsApp thread, with the screenshot and the product URL already on it.
bad-result.png
2 links
Where this came from
A price on the dashboard was wrong — the scraper had picked a junk first result. The CEO caught it and pinged the team on WhatsApp. An engineer read the thread, said it looked complicated and that they didn’t have the context, and forwarded it on. By the time it reached the CTO it had crossed three people and picked up no detail at all.
That company is PriceLantern, four people, and one of us is its CTO. Nowhere near needing Jira, well past the point where a tab full of issues is enough. They’ll be the first team running this.
09:04
09:04
Agent · over MCP
Read PL‑7, the attached URL, and the two scraper issues already open against that same store. issues.yourdomain.com/mcp
3 issues read1,180 tokens of context
09:06
Engineer
Hit Open in harness. The ranking bug came up already loaded as the prompt. Nobody had to say “which product, which store, since when” a second time.
09:31
Agent
Pushed pl-7-rank-fallback. Four commits, linked back to the issue on their own.
4 commits
!41 opened
09:33
Agent · WhatsApp
Posted the fix back into the thread the whole thing started in, so the person who spotted it heard without being chased.
10:05
Engineer
Reviewed the diff, merged, moved PL‑7 to closed. Three people, one thread, nothing explained twice.
Open
Planned
In progress
In review
Closed
What it does, all of it
No tiers, no seats, no roadmap page. This is the list.
Where it lives
Your domain, your database
One container at issues.yourdomain.com. Nothing phones home, and there is no hosted plan to get upsold into.
The work
Issues, files, people
Write an issue, attach files to it, assign the people who are actually on it. Sort a backlog and plan what lands next.
The code
Merge requests and commits
Link an MR or a commit and it stays linked both ways, so the issue always shows the code that answered it.
The public side
Two-way GitHub mirror
An issue here can keep a GitHub issue in sync, for the contributors who live over there.
For agents
MCP and an SDK
The board is a context source. An agent can read the backlog, the thread and the links before it writes a line.
Handoff
Open in your harness
One click sends an issue to your harness as a loaded prompt. Also reachable from the GitHub and WhatsApp agents.
Where it sits
The gap is real and we keep falling into it: a tab full of issues with no plan, or an install that needs its own admin.
GitHub Issues
Free and everywhere. No backlog, no plan, not yours.
issues
A backlog, links and an MCP. One container you own.
Jira
Every field you could want, and the meetings to fill them.
How you get it
Open source, with a hosted option for teams who would rather not run another box. The open one is the whole product, not a crippled tier.
Self-hosted
Free, and the real thing
One container at issues.yourdomain.com. Your database, your data, every feature on this page. No seats, no usage ceiling, no phone-home.
Hosted by Mirako
We run it and answer the phone
Same software, our instance, backups and upgrades handled. For four-person teams who have better things to do than babysit a container.
Planned — not priced yet. Paying for this is what funds the open one.Everything else in the shop
Labelled how they actually are. Mirako is a workshop, not a product line.
issues
Self‑hosted tracking with an MCP. The one we’re actually building.
no repo yet. It goes public when it runs
In development
lumiko
Generates documentation from a codebase with Claude, so what the agent reads is true.
mirako-studio/lumiko
Paused
toka
A harness where the shape of the solution matches the shape of the problem.
noxire-dev/toka
Paused
kodo
A web GUI for driving coding agents. T3 built the same idea faster than we could, so we stopped.
noxire-dev/kodo
Stopped
hako
Tools that didn’t want their own domain. Also where the Mirako design system lives.
internal, no repo
Internal
Why we’re building it
Every tracker on the market assumes a person is the only thing reading it. That assumption is now wrong, and bolting an MCP onto the side of one does not fix it.
So issues starts from the other end: one board, written and read by both hands, hosted where your code already lives.
Working on something near this?
Agent tooling, self‑hosted things, or you just want to argue about what an MCP surface should look like. We answer.
hi@mirako.computer