Kanera does not run your agents. Claude, Codex, and Cursor keep working wherever you already use them, and report here: they pick up cards, show live progress on the board, and flag when they need a person. Every change carries the agent's name, and there is no AI surcharge.
Agents already run in your terminal, your editor, and your chat app. Kanera gives them a shared place to report back: it writes down which agent acted and who it acted for, shows the work while it is still happening, and tells you about it afterwards.
Every comment, card change, and activity row an agent writes is recorded as the agent's, alongside the person it acted for. Feeds and comment headers carry a "via Claude" badge, so nothing an assistant did ever reads as somebody's own handiwork later.
Before it starts, an agent opens a run on the card. The tile carries a live chip naming the agent and the job, turns amber when the agent is waiting on a person, and clears when the run ends. A run that stops reporting is marked stalled rather than spinning forever.
Kanera does not notify you about your own actions — it makes an exception for your agent. What it changed while you were elsewhere lands in your inbox grouped as "you via Claude", so delegating work never means losing track of it.
An agent connected to Kanera works from live boards across every workspace your identity can reach, not from a stale export or a summary someone pasted in.
What changed on the Delivery boards this week?
Weekly status without asking anyoneThe agent reads activity across every board you can see and writes the summary you would have chased six people for.
What is overdue and who is blocked?
Triage what slippedRead whole blocking chains in one call, find upstream prerequisites and downstream work, and follow ready-to-start handoffs across accessible boards.
Prepare my standup from my Up next queue.
Standup, preparedYour ranked queue and assigned work, read the same way the app reads it, turned into three sentences.
Turn this thread into cards on the Support board.
Capture without retypingThe agent already has the conversation. It writes the cards, the checklists, and the labels in one pass.
Which cards reference the auth migration?
Find things you half-rememberSearch across cards, comments, notes, and attachments, then follow the links rather than reporting a list.
Audit the release board before we cut 2.0.
Check the work before it shipsUnchecked checklist items, unassigned cards, stale columns. The pass nobody has time to do by hand.
Remind me to renew the venue insurance next week.
An inbox that needs no boardName no board and the task lands in your private Scratchpad inbox instead of a question about where it goes. Dates read the way you said them, in your time zone.
Attach the failing test log and a screenshot to DEV-938.
Evidence on the card, not in the chatAn agent on your machine uploads files straight to the card through a one-time link, so long logs and screenshots never squeeze through a message.
What does the error in the attached screenshot say?
Reads what the team attachedImages come back as images and logs as text, so the agent answers from the files already on the card.
All three reach the same work, and all three stay inside your permissions. Most people need only the middle one, and it takes about a minute.
Your coding agent is already in a terminal, working on the thing a card describes. One install later it can read that card, comment on it, and close it when the work lands — so the board keeps up with the branch without anyone stopping to update it. Useful for you too: see what is on your plate without leaving the editor.
kanera setup claude also teaches your agent how to use Kanera properly.
Approve Kanera inside Claude, ChatGPT, Codex, Copilot, Cursor, OpenCode, or Gemini CLI once, in a browser — no key to create, nothing to paste — and from then on you just ask. It sees exactly what you see and nothing you cannot, and you can disconnect it from your profile at any point.
Headless client with no browser? Device authorization is supported too.
Connect an agentFor the agent that runs every Monday morning, the bot inside your own product, or the script that does the reporting nobody wants to do. Typed, dependency-free, and MIT licensed, so it runs anywhere you deploy and raises no questions from a legal review.
Signature checks, retries, and paging through long lists are already handled.
SDK documentationAlmost every project tool bolted a chatbot into the corner, picked one model, and asked you to live with it. Twelve months later that choice looks dated and you are stuck with it. Kanera built the doors instead, so the AI on the other side is always the one you rate today — not the one we happened to sign with.
You have a favourite. Your designer has a different one, and your engineers have opinions about both. Nobody has to learn a second-rate assistant bolted into a project tool just to get their work updated — everyone brings what they are already fluent in.
Kanera speaks MCP, REST, and a plain command line — the standards agents already use. When the thing everyone is talking about next quarter arrives, you point it at the same address and carry on. No migration, and no waiting for us to add an integration.
Every time your agent gets smarter, faster, or picks up a new skill, your Kanera workflow gets it too — the same afternoon, at no cost to you. A bundled chatbot only improves when the company that bundled it decides to improve it.
Connect Claude, ChatGPT, Cursor, or the CLI on the Free plan and let it read and update your boards. No AI add-on, no per-agent fee, no token bill, just fair-use limits on how fast agents can work.
Pro, at $5 per user per month, gives agents higher limits and adds unattended service agents and workspace API keys that run without anyone signed in. Host Kanera yourself and it is all switched on, with no plan checks anywhere in it.
Connect an agent in a browser and you are done. We run the address it talks to, keep it up, and keep it current. Interactive agents work on Free; unattended agents need Pro.
Run Kanera on your own hardware and everything here comes with it — your own agent address, the same tools, no plan checks, and nothing leaving your network that you did not send.
Install the Kanera plugin in ChatGPT or supported Codex surfaces, then connect through OAuth. Other MCP-compatible clients can use the hosted server directly. Kanera publishes setup guides for Claude, ChatGPT, Codex, GitHub Copilot, Cursor, OpenCode, and Gemini CLI, and the hosted MCP endpoint works with anything else that speaks the protocol.
Agents that can only run shell commands do not need MCP at all. They can use the Kanera CLI, which exposes every MCP tool as a command.
Yes. The official kanera command-line interface is published on npm as @kanera/cli and needs Node 22 or newer. Install it with npm install --global @kanera/cli, or run npx @kanera/cli commands without installing.
It exposes the same tool layer as the MCP server, prints a machine-readable command catalog, and uses exit codes that tell an agent whether a write was refused or failed. Running kanera setup claude installs the Kanera Agent Skill for Claude Code.
No. In ChatGPT, install the Kanera plugin and connect your account through a browser consent screen. Claude and Copilot also connect through OAuth.
The CLI and SDK are there for the cases where you do want to build something, and the API reference documents every endpoint if you would rather work directly against HTTP.
No, and that is the point. The best AI changes every few months, and a chatbot bolted into a project tool is frozen at whatever its vendor picked. Kanera is built to be reachable instead: a native MCP server, a CLI, a TypeScript SDK, a complete REST API, and a published Agent Skill.
So you use the assistant you are already fluent in, your colleague uses theirs, and when something better lands next quarter you point it at the same address and carry on.
Use browser-based OAuth for a personal assistant when the client supports it. Use device authorization for a CLI or SSH session that cannot open a browser redirect. Use an unattended service connection for CI, cron, or server-side automation, and an API key for local clients or those that only accept a static credential.
Exactly what the credential can see, and nothing more. A personal connection follows your current board access across workspaces, standalone boards, and guest boards, including your private personal notes.
An unattended service connection stays scoped to the single workspace or standalone board where it was created.
Yes. Create the personal API key as Read-only. That credential can list, read, search, and report, and Kanera refuses every write with a 403 at the API.
This matters because the refusal does not depend on the AI client honouring its own tool permissions. A confused or prompt-injected agent still cannot change a card, because the server will not let it.
Changes made through the API, MCP, or the CLI are recorded in card history like any other change, so you can see what happened and when.
For anything broad or sensitive, ask the agent to show a plan first, and give it a read-only credential until you trust it.
Yes. Connecting your own AI agent over MCP or the CLI, and personal API keys, work on every plan. Free boards have fair-use limits on how fast agents can work; Pro boards allow more.
Unattended service agents and workspace API keys need Pro, the 30-day Pro trial, or self-hosted Kanera, which has no plan checks at all. If a hosted organisation moves to Free, Kanera revokes its workspace API keys rather than deleting them; personal keys and interactive agent connections keep working. Upgrading again restores what was switched off.
Not by Kanera. Kanera never sends your work to a model provider, because Kanera does not call one.
When you connect an agent, your data goes wherever that agent sends it, under that vendor's terms. That is your choice to make and your relationship to manage, which is the point of keeping AI out of the product.
Yes. Self-hosted Kanera exposes its own MCP address, its own API, and works with the same CLI and SDK. Point them at your deployment and no request leaves your network except the ones your agent makes to its own model provider.
Connect in a browser with OAuth, or hand a read-only key to something that runs unattended. Either way, your work stays in Kanera and your choice of model stays yours.
