MCP servers for to-do apps

Linear MCP server

Yes. Linear hosts an official MCP server at mcp.linear.app/mcp. It signs in with OAuth 2.1 or a bearer token, and it's one of the few task tools with a separate read-only endpoint, mcp.linear.app/mcp/readonly, that exposes read tools only.

Official serverServer URLhttps://mcp.linear.app/mcp

How to connect it

  1. Claude: open Settings → Connectors and add the Linear connector (on Team and Enterprise, an owner can connect it for the organization).
  2. Claude Code: run the command below, then run /mcp to sign in.
  3. Codex: run codex mcp add linear --url https://mcp.linear.app/mcp, then codex mcp login linear if you configured it by file.
  4. Cursor installs it in one click from Linear's page; VS Code, Windsurf and Zed reach the same URL through npx -y mcp-remote.
claude mcp add --transport http linear-server https://mcp.linear.app/mcp

What it can do

  • Find, create and update issues, projects and comments
  • Work across issue relationships, milestones and cycles, which Linear's own example prompts use
  • Run read-only: through mcp.linear.app/mcp/readonly, the read OAuth scope, or an API key with only the Read permission
  • Act as an app user when a client passes a Linear OAuth token in an Authorization header

What it doesn't do

  • Linear's MCP page doesn't publish a full tool list, and says more functionality is coming.
  • Its docs don't describe an approval step on the issue itself; on Linear, review of agent work usually happens in the pull request.
  • Several Linear workspaces need separate sign-in contexts, set up client by client.
  • The SSE address, mcp.linear.app/sse, is deprecated; use it only where nothing else works.

Staying on Linear?

If you want an agent to report on Linear without touching it, point it at the read-only endpoint. It's the cleanest least-privilege setup any task tool here offers, and it takes no extra configuration.

When Done makes sense instead

Done is a to-do app built for handing work to an AI agent, with the parts an agent needs before you trust it with real tasks:

  • An approval gate: by default an agent can propose that a task is finished, but only you can close it. Done enforces that, so it holds even if an agent tries to skip it.
  • Questions instead of guesses: an agent parks a decision with its recommendation, and you answer in one click.
  • Plans before work: an agent can propose a plan that waits for your yes before it starts.
  • Lanes: name your agents and pin each task to one, so two agents don't pick up the same work.
  • Scoped access: tokens that are read-only or limited to one project or workspace, and budgets an agent can't raise.
How Done's MCP server worksDone vs Linear, feature by feature

Moving takes one step

In Linear, press Cmd/Ctrl+K and choose Export issues as CSV… from a view (members can export 250 issues at a time, admins 2,000). Done reads any CSV with a title column.

Done's import preview shows exactly what lands — projects, tasks, due dates, priorities — before you create an account, and nothing is uploaded.

Preview your Linear import

Other to-do apps

Try Done with the agent you already use

Free to start. Connect Claude, ChatGPT or any MCP client in about a minute.

Open Done

Sources

Information about Linear current as of October 2026, checked against the sources above. MCP servers change quickly, so check them before you rely on a detail. Linear is a trademark of its owner.