AI assistants

KanbanTrack for people and AI assistants

Understand the boards, sharing, and permissions, then connect an AI assistant to the projects you choose.

Get started

What KanbanTrack does

KanbanTrack helps freelancers and solo builders organize work, keep clients updated, and collect product feedback. Owners sign in with Google or GitHub and manage projects on simple boards. AI assistance supports these workflows when the owner grants access.

Choose a project type

  • Client delivery (client): plan tasks and share progress with a client. Statuses: To do (todo), In progress (in_progress), Waiting for client (waiting_client), and Done (done).
  • Product feedback (web_app): collect feature requests, bug reports, votes, and discussions from product visitors. Statuses: Open (open), In progress (in_progress), and Done (done). Feedback types are feature and bug.

Start with the project type that fits your work. Sharing, widgets, and AI connections are optional; you do not need every feature.

Sharing and website widgets

Clients and visitors do not need accounts. Anyone holding an active public link can see the shared card information and discussions. Client boards show progress and allow comments; product feedback boards also accept submissions and votes. Management remains controlled by the owner and any explicitly authorized AI connection.

Owners can rotate or revoke public links. Website widgets are for product feedback projects: configure your website and paste the generated script into it. Inline, floating, and custom-button display modes are available. Revoking or rotating widget access stops the old installation from working.

Done product feedback cards have read-only discussions; reopening the card restores replies. AI replies follow the same rules and carry owner attribution.

Connect an AI assistant

The remote MCP endpoint is https://kanbantrack.app/api/v1/mcp. Use Streamable HTTP. Setup depends on whether your agent app supports remote MCP and OAuth or custom bearer headers. Pasting a prompt can help with setup; it does not automatically authorize access.

  1. Add the MCP endpoint in your agent app’s connection settings, or paste the connection prompt and ask the assistant to guide you through manual configuration.
  2. For OAuth, follow the sign-in link, sign in with Google or GitHub, and approve selected projects and permissions. Start read-only. Access tokens last one hour; refresh tokens last 30 days.
  3. For apps or scripts supporting custom headers, open AI connections at /settings/ai, choose projects and permissions, and create a personal token. Configure Authorization: Bearer <token> in the client’s private settings. The secret is shown once, only its hash is stored, and tokens expire after 90 days. Keep it out of prompts and public documents.
  4. Ask the assistant to list your authorized projects, then choose one to work on. Reduce permissions or revoke the connection from AI connections whenever you need to. Expanding OAuth access requires renewed consent.

Connections may access selected projects or all owned projects, including future projects. Every call checks the credential, current connection permissions, and project ownership. Revocation blocks subsequent calls, including previously issued OAuth tokens. Browser sessions and public share links cannot authenticate MCP calls. OAuth access tokens must target this MCP endpoint.

For MCP client implementers: protected-resource discovery is at https://kanbantrack.app/.well-known/oauth-protected-resource/api/v1/mcp and authorization-server discovery at https://kanbantrack.app/.well-known/oauth-authorization-server/api/auth. Public clients register at /api/auth/oauth2/register with token_endpoint_auth_method set to none. Use authorization-code flow with PKCE S256, request kanban:read plus the permissions needed, and include offline_access for refresh. Set resource to https://kanbantrack.app/api/v1/mcp during authorization and token exchange.

Copy, paste into your assistant, then approve access. Setup depends on the agent app.

Four permissions

  • Read (kanban:read): Read authorized projects, cards, and discussions. This is the default and is required with other permissions. Reading discussions does not mark notifications as read.
  • Manage cards (kanban:manage): Create, edit, move, reorder, and archive cards in authorized projects. Includes reading. Does not grant project creation or permission to post replies.
  • Create projects (kanban:create-projects): Create client or product feedback projects. A project created by the connection automatically joins its selected-project grant. Separate opt-in.
  • Post replies (kanban:reply): Post comments as the project owner. Replies are visible to clients or feedback visitors with access to the shared board or widget. Separate opt-in.

Eleven MCP tools

  • list_projects: List authorized project summaries.
  • get_project: Read one project summary, counts, and valid statuses.
  • create_project: Create a client or product feedback project.
  • list_items: List cards in a project, with text, status, and feedback-type filters.
  • get_item: Read one card.
  • create_item: Create a card.
  • update_item: Edit supplied card fields, preserving omitted fields.
  • move_item: Move a card to the top of a status column or before/after another card.
  • archive_item: Archive a card.
  • list_comments: Read a card’s discussion without marking notifications read.
  • create_comment: Post a reply as the project owner, following the existing discussion rules.

Working with projects, cards, and retries

  • Use the project UUID as projectId and the card’s immutable issue label, such as KAN-42, as issueId. Resolve the label inside the authorized project. Use returned valid statuses for that project type.
  • Project summaries and card lists are separate. Paginated lists default to 50 results and allow at most 100 per page; use the returned cursor for the next page. Archived cards are excluded from active card lists.
  • create_project, create_item, and create_comment require a caller-generated UUID requestId. Generate one per intended creation and reuse it for a retry of the same request. Successful results are retained for 24 hours to prevent duplicate writes.
  • Card edits are partial. Move cards to the top of a column by omitting beforeIssueId and afterIssueId, or supply one neighboring issue label. Supply only one of these ordering options.
  • Validation and domain errors are structured tool results. Authentication failures are HTTP errors. Per connection, the server allows 120 calls and 30 mutations per minute.

Current limits

MCP does not support editing or archiving projects, managing share links or widgets, voting, managing notifications, or teams. Agent compatibility depends on the app; this guide does not promise support for every assistant or automatic connection.

This is public product and setup information. It contains no private project data or credentials. These HTML and text references are convenient ways for assistants to understand KanbanTrack; they do not guarantee AI-search visibility.

Start with one project.

Sign in with Google or GitHub, create your board, and choose what to share.

Create your board