List tasks
The project’s columns and their tasks. By default, those not in a done column; also by column.
Kanlane with Claude
Connect Claude to your tasks with the Kanlane MCP server
Give your assistant a token for one project and it can see its tasks, move them to “In progress” or “Done”, leave a note about what it did and create new tasks. You decide which project, whether it only reads, and when it ends.
Updated:
In short
Kanlane has an MCP (Model Context Protocol) server: an address an AI assistant, such as Claude Code, connects to in order to work with the tasks of a project. It works with a token you create in Settings, it is valid for a single project, it can be read-only and you can revoke it whenever you want. Kanlane does not include an assistant of its own and does not call any model: you connect yours. It does not work with encrypted projects, and the assistant cannot see your client records, your meetings or your passwords.
What the assistant can do
Five tools, all about the tasks of the token’s project. They work with the real columns of your board, whatever they are called.
Read a task
Description, subtasks, date, labels and its latest notes and activity lines.
Move between columns
From pending to in progress, or to a done column to complete it. The task goes to the end of the column.
Add a note
To say what was done or what is left. Plain text or Markdown.
Create a task
With title, description, column, due date and labels.
Recurring tasks
When one is completed, the next one is created, just as when you complete it yourself.
What it looks like on the task
An example with made-up data. What an assistant does is told apart from what you do: it is signed with the name of whoever created the token and the name of the token.
L
Build the new landing pageIn progress · North Studio
ActivityMoved from “To do” to “In progress” by Ana · MCP (token “Laptop”).
Note by Ana · MCPMock-up pushed to the staging branch. Mobile still needs a review.
TokenLaptop · read and write
Last used3 min ago
How to connect it
- Open the project and go to Settings → Integrations. The section is called “AI assistants (MCP)”. You need an account: it is not available in guest mode.
- Create a token. Give it a name you will recognise and tick “Read-only” if you do not want it to change anything.
- Copy the command and paste it in a terminal. The token is shown only once. Kanlane does not store it: it only keeps its fingerprint to recognise it.
- Ask your assistant. For example: “list the pending tasks in Kanlane”, “move this one to In progress” or “mark it as done and leave a note with what you changed”.
The command for Claude Code
Kanlane gives it to you already written with your token. It looks like this:
claude mcp add --transport http kanlane https://kanlane.com/__/mcp/v1 --header "Authorization: Bearer kl_…"Other MCP clients connect to the same address, https://kanlane.com/__/mcp/v1, over HTTP and with the token in the Authorization header.
Permissions and security
What you control
- Each token is valid for a single project. It does not even work for another project in your own account.
- It can be read-only: then the assistant does not even see the tools that write.
- You revoke it from Settings and it stops working on the next call.
- In a team, the token can do what you can do at that moment: if you become a viewer, it stops writing; if you leave the team, it stops working.
- Every change is recorded in the task’s activity, with your name and the token’s name.
What you should know
- What the assistant reads leaves Kanlane for your assistant’s service, under its own privacy terms.
- Whoever has the token can use it: treat it like a password.
- A task may contain text written by someone else or received by email. Kanlane hands it to the assistant marked as data, not as instructions, but you decide what you let it do.
- The “when created” and “when moved” automation rules are not triggered by what the assistant does.
- There are limits: 60 calls per minute per token and 2,000 changes per day per project.
This is not for you if…
- Your project is encrypted, with “Full encryption” or “Managed by Kanlane”: the server does not have its key and cannot read its tasks.
- You expect an assistant inside Kanlane: there is none. Kanlane does not call any model; it only opens the door to yours.
- You want the assistant to look up your client records, the calendar or the passwords: it only works with tasks.
- You need it to edit the title or description of an existing task, or to delete it: there is no tool for that.
- You want your automations to run when the assistant moves or creates a task.
- You use Kanlane as a guest or in local mode: without an account there are no tokens.
- You only use the claude.ai web connector: it has not been tested and may require a kind of access Kanlane does not offer yet.
Frequently asked questions
What is the Kanlane MCP server?
It is a Kanlane address that an AI assistant compatible with the MCP (Model Context Protocol), such as Claude Code, connects to. With a token for your project, the assistant can read its tasks, move them between columns, add notes to them and create new tasks.
What can the assistant see and change?
Only the tasks of the project you created the token for: title, description, column, dates, labels, subtasks, notes and the name of the client and contact assigned to each task. It cannot see the client and contact records, the meetings or the passwords, and it cannot delete tasks, change the columns or touch the project’s permissions.
Can I give read-only access?
Yes. When you create the token you tick “Read-only” and the assistant can list and read tasks, but not move them, annotate them or create them. If you can only read the project in a team, your tokens are always read-only.
Does it work with encrypted projects?
No. In a project with “Full encryption” or “Managed by Kanlane” the server does not have the project’s key, so it cannot read or write its tasks. If a project becomes encrypted, its tokens are removed.
Are my automations triggered when the assistant moves a task?
No. The “when created” and “when moved” rules are run by the browser of whoever changes the task, and here the change is made by the server. The only exception is a recurring task: when it is completed, the next one is created.
Which assistants does it work with?
With MCP clients that connect over HTTP and send the token in the Authorization header. It has been tested with the official client of the protocol, and the command on this page is the one for Claude Code. The claude.ai web connector has not been tested and may require a kind of access (OAuth) that Kanlane does not offer yet.
What do I do if I lose a token or no longer need it?
Revoke it in Settings → Integrations. It stops working on the next call and cannot be recovered; if needed, you create another one. Deleting a project removes its tokens, and deleting your account removes all the ones you created.
Does it cost anything or have limits?
There is no extra cost. Each token allows 60 calls per minute and each project 2,000 changes per day across all its tokens; reading does not count towards that cap. A project can have up to 10 tokens.
Try it on a project
Create your account, open a project without encryption and go to Settings → Integrations to create the first token. The demo shows the board with sample data, but it has no MCP server: you need an account.