Use kanman from Claude Code or Codex

You can use kanman from your own Claude Code or Codex. One connection covers your whole workspace: every team, every story you can see in kanman. You sign in once in the browser; there is no key to copy and nothing to set up per team or per story.

Your agent then works with kanman the way you do in the app, with your role:

• take the next story, work on it in your own checkout and hand in the change for checking;
• see stories and decisions, and answer them;
• ask kanman, as in your conversation in the app.

Everything it changes is recorded as you.

Connect Claude Code

Run this once, in any terminal:

claude mcp add --transport http --scope user kanman https://api.kanman.ai/functions/v1/mcp

Then start Claude Code, type /mcp, choose kanman and Authenticate. Your browser opens:

1. Sign in to kanman, if you are not signed in yet.
2. kanman asks: Allow Claude Code to work in kanman as you? It shows your account and the workspace. If you belong to several workspaces, pick the one this connection is for.
3. Click Allow. The browser goes back to Claude Code, and the kanman tools appear.

Claude Code keeps the connection and renews it on its own. You do not need to sign in again unless you disconnect it.

The command is also in the app: open a team’s settings, Agent and models, and find On your machine. A story page links there too (Work on it in Claude Code).

Connect Codex

codex mcp add kanman --url https://api.kanman.ai/functions/v1/mcp
codex mcp login kanman

codex mcp login opens the browser for the same sign-in. Then start Codex in your repository.

Which team kanman works on

kanman picks the team from the repository you are in. Your agent asks kanman who you are and passes the repository’s address (git remote get-url origin); kanman answers with the team that works on that repository. If two teams work on the same repository, kanman lists both and your agent asks you which one you mean.

You can always name the team instead (“work on the payments team”), or a story key such as PAY-12: a story key finds its own team.

Work on the next story

In your repository, ask Claude for the next story, for example:

Take the next story from kanman, implement it in this repository and commit it. Then submit it for review with the branch and the diff (git diff main...HEAD).

What happens:

1. Claude gets the next story from the team’s Ready column (or the story you named) and moves it to In progress.
2. Claude implements it in your checkout, with your own git access. On the way it reports progress, can ask the team a question (it appears under Decisions) and checks the team’s policy before it touches sensitive paths.
3. When the work is committed, Claude submits the branch, the diff and, if you opened one, the merge or pull request link. Pushing and opening the merge request stay with you.
4. kanman checks the change against the acceptance criteria and the team’s policy limits, such as denied paths and the size of the change, and posts the evidence on the story.
5. If something is missing, Claude sees what to fix the next time it looks at the story, fixes it and submits again.
6. When everything passed, Claude completes the story and it moves on to Review or Done.

kanman checks your change with the same checks as its own work and holds it to the same standard. While your merge request is open, kanman starts none of its own stories on the files it changes. If one of its runs reaches those files anyway, the run pauses until your change has landed and then continues on top of it. The story’s card says which change it waits for.

On a team without a repository, no acceptance spec runs, and the evidence says that the outcome is not proven. See Use kanman without a tracker.

What your agent can do

Your agent can do what you can do in kanman, never more. A viewer can read but not change anything; members and admins can work on stories. The tools and their inputs are listed in the MCP tools reference.

What needs me? On teams where kanman handles the questions itself, ask your agent this and it shows kanman’s brief right away: each question with kanman’s recommendation and what follows from each answer. Tell your agent your choice and it answers the item for you; the answer applies to every story the item covers and shows in the app, Slack and Teams within seconds. Why did kanman approve INT-241? gets the answer from what kanman recorded at the time.

Text that kanman passes on from others, such as a story’s description from a tracker or a CI log, is marked as data for your agent, so instructions hidden in it are not followed.

Disconnect

Open Your settings > API tokens (avatar menu). Under Connected apps you see every app that works in kanman as you, with its workspace and when it was last used. Click Disconnect to stop it at once. To use it again, sign in from it once more.

CI and scripts

Where nobody can sign in in a browser, use a workspace API token instead:

1. Create a token in Your settings > API tokens for the workspace.
2. Store it as a secret, for example KANMAN_API_KEY, and pass it as a header:

claude mcp add --transport http kanman https://api.kanman.ai/functions/v1/mcp \
  --header "Authorization: Bearer $KANMAN_API_KEY"

A token covers working on stories. Answering decisions, talking to kanman and the other tools that act as you need the browser sign-in.

Related

• MCP tools reference
• Use kanman without a tracker
• Decisions