Kaneo and the twenty-sixth seat

Hire your twenty-sixth person and your Jira bill doubles the same week, for exactly the product you already had. Atlassian bills annual plans in bands rather than per head, and its own licensing page puts the 16-to-25 band at 2,250 dollars a year and the 26-to-50 band at 4,550. No new features arrive. Nothing gets faster. One person joined, and the price of tracking work went up by 2,300 dollars. Kaneo is an MIT-licensed project tracker that runs the same boards, statuses, assignees and labels on a twelve-dollar server, it crossed 7,400 stars this week, and yesterday it shipped the part that actually changes the job: an MCP server, so the person who runs your operations can create and move tickets by asking, instead of filing a request for someone to build that integration and then waiting a quarter.

The band steps, it does not scale

Most people assume seat pricing is linear. You pay per person, you add a person, you pay a bit more. That is roughly true on monthly billing and it is not true at all on annual, which is the plan most small companies pick because somebody told them annual is cheaper.

Here is what the tiers actually say. Jira Standard on annual billing: 900 dollars for up to 10 users, 1,350 for 11 to 15, 2,250 for 16 to 25, 4,550 for 26 to 50, 9,050 for 51 to 100. You are not buying seats. You are buying a band, and you pay for the whole band whether you fill it or not. A company with 26 people pays the same as a company with 50.

Now run the other side of it. Jira Standard on monthly billing is 9.05 dollars per user for the first 100 users. Twenty-six people at 9.05 for twelve months is 2,823 dollars and change. Annual billing for those same 26 people is 4,550. The "cheaper" plan costs 1,700 dollars more.

That inversion holds until roughly 42 users, where the flat 4,550 finally beats paying by the head. So if your company sits anywhere between 26 and about 41 people and you are on annual Jira, you are paying a premium for the privilege of prepaying. Go look at your renewal. This is the least glamorous money any operations lead will find this month and it takes four minutes.

I am not telling you that to sell you on a repository. I am telling you because it is the honest first move, and because it frames what the open source option is really worth. The question is never "free versus 4,550." It is "what am I actually buying for 4,550, and is it that."

What Kaneo actually is

A board. Deliberately, almost aggressively, just a board.

Kaneo is a React and Tailwind front end over a Hono API and a PostgreSQL database, and the project's own documentation is refreshingly short about it. You get workspaces, projects, tasks, a board view and a list view over the same source of truth, statuses, priorities, labels, assignees, due dates, comments, task relations, roles and permissions, and a native GitHub integration that syncs issues. The founder, Andrej Acevski, is explicit on the homepage that the absence of everything else is the product. His framing is that the problem with the incumbents was never missing features, it was too many of them.

The traction is real and it is recent. The repository shipped v2.12.2 on August 4, roughly ten tagged releases landed between July 27 and August 4, and the last commit is a day old. The LICENSE file is plain MIT, copyright 2024 Andrej Acevski, which means commercial use, modification and redistribution with no strings and no source-available asterisk. That matters more than it sounds. Several of the self-hosted tools in this category are AGPL, which is fine for internal use and gets complicated the moment you want to put a modified version in front of customers. Kaneo has none of that friction.

There is also a hosted option, and it deserves saying plainly rather than burying at the bottom: you do not have to touch a server at all. Kaneo Cloud is 50 dollars per user per year on the Team plan, 40 dollars a year flat for a single user on Personal, with a 14-day trial and no card up front. Twenty-six people on Kaneo Cloud is 1,300 dollars a year against Jira's 4,550. Self-hosting is free forever and, notably, includes every feature including SSO, which is the exact thing most open core products hold back as the upgrade bait.

The part the coverage is missing

Every write-up of this repo I have seen files it under "open source Jira alternative" and stops. That is the boring half.

The interesting half shipped yesterday. Kaneo published @kaneo/mcp to the official MCP registry, and the server is also built into the app at an /api/mcp endpoint. Read the tool list and you can see the whole shape of it: list_workspaces, list_projects, create_project, list_tasks, create_task, update_task, move_task, update_task_status, create_task_comment, create_label, attach_label_to_task, create_task_relation. That is not a read-only reporting connector. That is write access to the tracker, exposed to whatever assistant you already have open.

Think about what that replaces for the operations reader specifically. The request you cannot get prioritized is almost never "build us a project management tool." It is the seam: when a deal closes, open the onboarding project with the seven standard tasks, assign three of them, label them by account tier, and link them to the renewal task. Today that is a ticket, and the ticket sits behind whatever the product roadmap says is more important, which is everything. With an MCP server pointed at your own instance, that is a sentence you type, and the assistant does it against a real API with real permissions.

Setup is npx @kaneo/mcp, which runs an interactive installer that registers the server into Claude Desktop or Cursor for you. You need Node 20 or better, a running Kaneo instance, and a KANEO_API_URL. For anything unattended you set KANEO_API_KEY instead of going through the device login flow. That is genuinely a ten-minute job for someone who has never opened a terminal for anything more serious than a Wi-Fi reset, and it is the first time I have seen the write side of a tracker handed to a non-engineer this cheaply.

Worth noting the contrast, since it is the whole argument for owning the thing: on the commercial side, agent and MCP access is a line in the feature matrix, which means it is a thing a vendor can move. Linear's pricing page lists MCP access and its agent platform in the same table as the features that separate the 10-dollar plan from the 16-dollar one. When you host the tracker yourself, the API is just the API, and nobody is deciding which tier of customer gets to automate.

What it actually costs to run

This is where "free" stops being a useful word, so let me put real numbers on it.

The fast path is drim, a CLI the project maintains: one curl to install, then drim setup, and it handles the compose file, the database and HTTPS. The manual path is Docker Compose with a Postgres service, a .env holding POSTGRES_PASSWORD and an AUTH_SECRET you generate with openssl rand -hex 32, and a KANEO_CLIENT_URL pointing at wherever it will live. Then docker compose up -d.

Honestly: three to four hours the first time, for a competent non-engineer who has never done this before. Maybe ninety minutes if you have stood up a container before. The compose part is twenty minutes. The rest is the stuff nobody puts in a quickstart, which is a domain name, a TLS certificate, deciding whether this thing faces the public internet or sits behind a VPN, and setting up a database backup that you have actually tested restoring rather than merely configured.

Hosting is a twelve-dollar DigitalOcean droplet at 2 GiB of memory and 50 GiB of disk, which is comfortable for a team of this size running an app and a Postgres alongside it. Add backups, priced at 20 percent of the droplet for weekly or 30 percent for daily, and you are at roughly 15 dollars a month. Call it 180 dollars a year against 4,550, plus the four hours, plus whatever your time is worth on the day something breaks.

And something will need attention, because ten releases landed in nine days. That cadence is a good sign about the project and a real ongoing cost for you. Self-hosting means you own the upgrade treadmill. Nobody patches your instance on a Tuesday night while you sleep.

The honest take

Three things I would want said out loud before anyone migrates.

The first is the bus factor, and it is the same one that shows up in almost every repo in this category. Kaneo is overwhelmingly one person's work, with dependency bots and a release bot filling in the commit log around them, plus a handful of named sponsors funding development time. The mitigation is better than usual, because MIT plus your own Postgres means you keep running the version you have indefinitely if the project stops, and you can pay someone to patch it. But be clear-eyed: the reason the release cadence is this fast is the same reason it could go to zero without warning.

The second is that the commercial side is brand new. Kaneo Cloud's paid plans launched recently enough that the pricing page is still explaining how existing free accounts get grandfathered, promising at least twelve months of continued access and six months of notice. I read that as a founder trying to be straight with early users, which I respect. I also read it as a business model that has not yet been tested by a full renewal cycle. If you pick Cloud over self-hosting, you are betting on a revenue line that is younger than most of the tasks you would put in it.

The third is that the minimalism cuts both directions, and this is the part that decides the whole thing. Jira Standard is not 4,550 dollars of nothing. It is timeline and calendar views, dashboards and reporting, capacity planning, cross-project dependencies, approvals, 1,700 automation rule runs a month, 250 GB of storage, mobile apps, and a support contract with a human at the end of it. Kaneo's own marketing does not claim most of that, and you should take it at its word rather than assume the gap will close. If your team lives in Jira dashboards, or the finance lead pulls a burn report out of it every month, or field staff update tickets from a phone in a truck, this is not a swap. It is a downgrade you would notice in week two.

But if your actual usage is what most small companies' usage actually is, a board with columns, people assigned to cards, and a due date, then you are paying enterprise prices for the ten percent of the product you touch. That is the trade. Not "open source good, vendor bad." Just: which one are you.

One more thing on data, since it is the question every operator should ask and almost none do. Self-hosted, the tasks sit in a Postgres database on a machine you control, and the only thing leaving the building is whatever you deliberately point at an outside service. The MCP server changes that calculus slightly, because the moment an assistant reads your tickets, your project names and your comment threads are going to a model provider. That may be entirely fine. It is not nothing, and it deserves a decision rather than a shrug.

The pricing band is the thing I keep coming back to. Somewhere out there is a company that delayed a hire, or split one across two systems, or quietly told a contractor to share a login, because crossing from twenty-five people to twenty-six was going to cost 2,300 dollars in tracking software. That is a pricing model reaching into a staffing decision, and once you have seen it you cannot unsee it on the rest of your stack.