Your projects, tasks and coding agent inside VS Code

Manage a backlog without leaving VS Code

The usual arrangement puts the task manager in a browser tab and leaves the editor knowing nothing about it. Status then depends on somebody remembering to switch tabs and transcribe what they did, which is why most backlogs describe the repository as it was a fortnight ago.

Putting the backlog in the editor only helps if the thing doing the work can write to it. That is what the language-model tools are for: a session that finishes a task marks it completed and leaves a comment saying what it verified, and the node recolours in the map without anyone opening a tracker.

This holds without Autoplans in the picture. Any arrangement where the tool writing the code can also record what it finished will keep a truer backlog than one that asks a person to copy progress across.

GitHub Copilot project management from one backlog

Copilot's agent mode can call the language-model tools any installed extension contributes. Autoplans contributes nineteen, so a Copilot session can list your projects, read a task with its subtasks, create the follow-up work it discovers, and set statuses as it goes — against the same account the dashboard shows.

The instructions file matters as much as the tools. Autoplans: Generate Copilot Configuration asks the editor for a model, reads what the repository already documents — .github/copilot-instructions.md, AGENT.md, AGENTS.md, CLAUDE.md, .cursorrules, .windsurfrules, .clinerules, README.md — and writes an instructions file that keeps those conventions and names the task tools. Without a model available it falls back to a template. There is more on writing that file by hand in the guide to Copilot instructions at /guides/copilot-instructions.

The task map in VS Code

A node is a task. Border colour is status: emerald for completed, cyan for in progress or waiting for review, amber for blocked or failed, grey for pending. An amber dot marks high and critical priority, and a line of counts under the title gives the priority, the subtasks and the comments.

Arrows are dependencies and run left to right, so a task sits to the right of everything it depends on. The hub on the left is the project; tasks with no dependencies connect to it, which makes the leftmost column the work that can start now.

Clicking a node opens the drawer: description, status, priority and type, subtasks, relations you can click to jump to, linked commits with their branch and line counts, and comments with agent comments labelled. Start, Mark done and Reopen act on the task without leaving the panel.

A VS Code extension with an MCP server behind it

Everything the extension shows comes from the Autoplans MCP server at https://autoplans.dev/api/v1/mcp, over Streamable HTTP with your API key as a bearer token. The extension is one client of that endpoint; Claude Code, Cursor, Windsurf and Claude Desktop are others, and they all see the same projects.

What the extension adds on top is what a bare MCP connection cannot give you: the sidebar, the map as a panel, agent sessions, and the chat participant. Per-client configuration blocks are at /mcp-server, and the VS Code and Copilot one at /mcp-server/vscode-copilot.

You do not need the extension to give Copilot the backlog, though. VS Code reads MCP servers from .vscode/mcp.json in the workspace, and an input variable keeps the key out of the file you commit:

{
  "servers": {
    "autoplans": {
      "type": "http",
      "url": "https://autoplans.dev/api/v1/mcp",
      "headers": { "Authorization": "Bearer ${input:autoplansKey}" }
    }
  },
  "inputs": [
    { "id": "autoplansKey", "type": "promptString", "description": "Autoplans API key", "password": true }
  ]
}

A VS Code extension for AI sprint planning

Describe the project once — in the dashboard, to the chat participant, or in an agent session — and the planning agent returns tasks and subtasks with dependencies between them. The result is a draft to edit, not a plan to accept; renaming, deleting and reordering is expected.

There is no sprint object in Autoplans: no iteration, no burndown, no velocity chart. What planning produces is a dependency-ordered backlog, and the leftmost column of the map is what is unblocked today. If your team runs sprints, that column is what you pull from.

Related