Most people manage tasks by switching apps: you’re talking to Claude about a project, you think of three things to do, and you tab over to your to-do app to type them in. Then later you tab back. The list and the assistant never actually meet.
MCP fixes that. Once your to-do app is connected to Claude over the Model Context Protocol, you can add, find, and reschedule tasks from the conversation — and they land in your real list, synced to every device. Here’s how it works and how to set it up.
What MCP actually does
The Model Context Protocol is an open standard for letting AI clients talk to external tools. Instead of every app building a one-off Claude integration, any MCP client (Claude, Cursor, Codex) can connect to any MCP server. Your to-do app is the server; Claude is the client. Connect them once and Claude gets a live, read-write view of your tasks.
Concretely, that means you can say things like “add ‘renew domain’ to my Today list,” “what’s overdue in the Acme project,” or “push everything tagged errands to the weekend” — and Claude does it against your actual data, not a copy it forgets when the chat ends.
Connecting Lume to Claude
Lume ships a first-party MCP server, so there’s no community bridge to maintain. The setup is a few lines in your client’s MCP config. In Claude that’s an entry like this:
{
"mcpServers": {
"lume": {
"url": "https://lumetasks.com/api/mcp",
"headers": { "Authorization": "Bearer <your-token>" }
}
}
}Drop that into your MCP config (for Claude Desktop and Claude Code that’s the MCP settings file), restart the client, and Claude can now see your tasks. Each connection is scoped to your account with its own token, so the assistant only ever touches your own list, and you can revoke access at any time.
What it feels like day to day
The shift is subtle but real. Planning stops being a context switch. You finish a back-and-forth with Claude about a feature and just say “turn those into tasks under the Billing project” — done, no app-hopping. During a daily review you ask “what did I not get to yesterday?” and triage out loud. The list stops being a place you visit and becomes context the assistant already has.
One step further: let it do the tasks
Reading and writing your list is the table-stakes use of MCP. The more interesting one is hand-off. Because Lume also lets you assign tasks to Claude Code and Codex, you can go from “add this task” to “do this task” in the same breath — the agent picks it up, works in the background, and returns it for your review. The list isn’t just visible to your AI; it’s actionable.
Getting started
You’ll need a Lume account and the MCP server (it’s on the Pro plan). Add the config above, point it at your token, and ask Claude to list your tasks to confirm the connection. From there, managing your to-dos from a conversation just becomes the way you work.
More detail on the server itself is on the MCP task manager page, and Lume is free to start.