Labsco
MCP SERVER

MCP Kanban Memory

by eyalzh

Have the agent keep its plan on a kanban board with work-in-progress limits, so a long build survives across sessions.

Project & Task Management
Summary
Task memory the agent has to respect, not just write down.

Column capacity limits are what separates this from a to-do list the model keeps for itself: the board refuses when it tries to start a fifth thing at once. Because the database sits in a folder you named, "find the board for this project and carry on" still works a week later, and you can open the web UI and reorder it yourself in between.

What it is

An MCP server with an embedded SQLite database and a web UI. The agent records its plan as tasks on a board and moves them between columns as it works; you watch and edit in the browser.

What you get
  • create-kanban-board to open a board for a project
  • add-task-to-board and delete-task to keep the board matching the plan
  • move-task to advance a task, with column capacity — work-in-progress — limits enforced
  • get-board-info and get-task-info, which is how a later session finds a board and reads what was already done
  • State in an embedded SQLite database, so the board outlives the conversation
  • A web UI for watching progress and editing tasks by hand
  • Prompts for starting a project and for resuming one
Requirements

Node and a clone of the repository, built with npm across shared/db and mcp-server. One environment variable points at the folder holding the database files — it needs read and write access, and is created if it does not exist. A Docker image builds from the same repo — mount that folder at /mcp.