Everything you need to know about AgentKanban

AgentKanban is a task board built for working with AI coding agents inside VS Code. You plan work as boards, lanes, and tasks in the browser, then bind those tasks to your agent harness — GitHub Copilot or Claude — so each chat is captured against the work it belongs to. The board stays the source of truth for the work, and the agent conversation becomes resumable, shareable task history instead of a chat tab you eventually lose.

It is for developers and teams who run real work through AI coding agents and want more structure than an ephemeral chat window. That includes solo developers juggling several tasks, small teams sharing context, and larger organisations running many active projects who need the work, the reasoning, and the agent history kept together in one place.

Ephemeral chats work fine for small one-off tasks but break down on larger work and across a wider project. Useful reasoning, manual steering, and decisions vanish into a single chat tab, and resuming means dragging a bloated transcript forward. AgentKanban keeps the task, the conversation history, the steering, and the board-level context attached to the same unit of work — so you can stop, resume, or hand off without losing the thread.

Both a web app and a VS Code extension, working together. The hosted web app at agentkanban.io is where you create and collaborate on boards in real time. The VS Code extension connects your editor to a live board, lets you pick the active task from the sidebar, and captures agent harness turns against it. You can use the board on its own, but the agent context capture and resume features live in VS Code.

No. The boards, lanes, tasks, comments, todos, attachments, and team features work as a normal collaborative kanban tool. The agent integration is an additional layer that you opt into when you want your AI coding work captured and resumable.

AgentKanban works with GitHub Copilot and Claude (Claude Code) inside VS Code. You plan and steer on the board, then run the agent in whichever harness you prefer, with the active task providing context.

When you bind an agent harness chat to the active task in VS Code, prompts, responses, and your task-level steering are captured against that task. Instead of living in a single chat tab, each turn becomes part of the task timeline — so the useful parts of the execution stay attached to the work itself.

Use @kanban /resume to reopen a task in a fresh chat with task-specific context already injected. Rather than carrying a long, noisy transcript forward, the new session starts from the relevant history — the timeline, comments, todos, and board memory for that task. That keeps context lean and gives the agent a cleaner starting point.

On their own, the agent's memory is the chat window: close it and the reasoning is gone, and there is no shared place for the rest of your team to pick it up. AgentKanban stores that context on the task, so work survives across sessions and across people, and sits next to the board, comments, and decisions where the feature actually lives.

They are a structured way to move a task from intent to finished work: plan the approach, break it into an iteration-based todo list, then implement against it. Keeping those phases on the task means the plan and its reasoning are still there when the feature changes later.

AgentKanban ships a Model Context Protocol (MCP) server so MCP-compatible tools can read and update boards, tasks, comments, todos, assets, board memory, and technical docs directly. That lets your agent treat the board as a live data source rather than something you copy and paste into a prompt.

It lets agent work on a task happen in an isolated git worktree so parallel tasks don't collide in the same working tree. You get cleaner separation between concurrent pieces of work without manually juggling branches and checkouts.

They are durable, board-level context — project guidance, technical notes, and agent instructions (including shared AGENTS.md and memory notes) — that persist beyond any single task or chat. Every task starts with the right context already in place, so you are not re-explaining the project to the agent each time.

Yes. Boards sync in real time, so people can plan, comment, and reorganise work together. Because task context is stored on the task rather than in a private chat, another member can pick up where you left off and inherit the relevant reasoning, comments, todos, and notes.

You can invite members, manage roles, create teams, and switch between organisations. Boards can be shared with specific users, a team, or the whole organisation, so access matches how your projects are actually structured.

Captured task context is stored on the task, which means it is available to others with access to that board. That is the point — it turns private chat history into shared project context so work doesn't get stuck with one person.

Yes. You can upload files to tasks and comments, preview images, and copy portable asset tokens so agents can resolve those files later when they work on the task.

Yes. The free tier is built for evaluation, solo work, and small teams — it includes the full workflow with generous limits (for example, up to 3 organisation members and up to 3 active boards). See the pricing page for the current limits.

Pro raises the limits for organisations running multiple active projects at scale — many more boards and active tasks, higher API and MCP request allowances, far more attachment storage, and longer turn and activity retention. The full comparison is on the pricing page.

Pro is billed at the organisation level and is seat-based. Organisation owners purchase seats and assign them to invited members. You can increase or decrease your seat count, or cancel, at any time, and your subscription updates automatically. Payments are processed by Stripe.

Turn and activity retention is 14 days on the free tier and 365 days on Pro. Board memory, technical docs, comments, and todos are included on both plans. Exact limits can change, and notice of any reduction in plan limits is given at least 30 days in advance.

Yes. If you are unhappy with a purchase within 14 days of payment, contact us at [email protected] for a full refund.

Several tools now put a kanban surface in front of AI coding agents, and they often focus on orchestrating and running agents from a local board. AgentKanban's emphasis is different in three ways: it is VS Code-native, so the board lives next to where you already write code; it captures and resumes context on the task, so agent history becomes durable and lean rather than tied to a chat tab; and it is built for real-time team collaboration with hosted, shareable boards. If your priority is keeping task context and reasoning attached to the work and shared across a team, that is the gap AgentKanban is built to fill. The best way to judge fit is to try the free tier on a real task.

General project trackers manage tickets, but they don't know anything about your agent harness or what happened during an agent session. AgentKanban is purpose-built for that link: it binds the active task to your VS Code agent, captures the conversation against it, and lets you resume with the right context. You can think of it as the operating layer for agent work rather than a replacement for your company-wide tracker.

Yes. There is an external API and webhooks so you can automate board changes and sync with other tools, plus the built-in MCP server for agent access. That lets AgentKanban sit alongside your existing stack rather than forcing a migration.

Create a free account at agentkanban.io, make your first board and a task, then install the VS Code extension and connect it to that board. Pick the active task in the sidebar and start working with your agent harness. The Getting Started guide walks through it step by step.

From the Visual Studio Marketplace or the Open VSX Registry. Search for “AgentKanban” in the Extensions view, or use the links above.

The agent context capture and resume features are built around VS Code and VS Code-compatible editors that can run the extension (the extension is published to both the Marketplace and Open VSX). The hosted board itself runs in any modern browser, independent of your editor.

Your boards, tasks, and captured context are yours. Payments are handled by Stripe, and we aim for high availability across the hosted service. For specifics, see our Privacy Policy and Terms of Service.

Email [email protected] and a real human will get back to you, or read the documentation. You can also reach us from the contact page.