DoLoop has an MCP server. MCP (Model Context Protocol) is the standard way for an AI assistant to use tools from another app. Once Claude Code is connected, you can ask it what's on your list, have it fill the todo, and tell it "I've cleaned the sink" instead of opening the app to click Check.
This guide covers one client, Claude Code, using the browser sign-in. You don't create or paste an API key at any point. We ran every step below on 9 November 2026 with Claude Code 2.1.319 and a DoLoop Pro test account on our test server, which runs the same build as production.
Before you start
- DoLoop Pro. The API and the MCP server are part of Pro. A Free account can't connect: the sign-in page offers the upgrade instead of an Approve button, and the server refuses requests from Free accounts. The pricing page has the details.
- Claude Code, installed and working in a terminal.
- A few loops in DoLoop. The assistant works with what's already there.
Step 1: connect Claude Code
Run this in a terminal:
claude mcp add --transport http doloop https://api.getdoloop.com/mcpStart Claude Code, type /mcp, choose doloop, then Authenticate. Your browser opens a DoLoop page titled "Connect Claude Code to your DoLoop account". Sign in first if you aren't already.
Read that page before you approve. It shows:
- An Unverified app badge. Claude Code identifies itself with a document hosted on claude.ai, and the page shows that host. DoLoop keeps no list of approved apps, so every app gets this badge, and the name is the app's own claim. Only continue if you started the connection yourself.
- What the app will be able to do. By default it asks for full access: view, create, change, check, skip and delete your tasks and loops.
- A Read-only access option. Tick it if you only want the assistant to look. It then gets 5 read tools instead of all 16.
Click Approve. Back in the terminal, Claude Code reports "Authentication successful. Connected to doloop."
Step 2: start by reading
Your first request should change nothing:
What's on my DoLoop list today, and which loops do I have? Don't change anything.Claude Code calls two read tools: one for today's tasks and one for your loops. In our test, the list was empty because nobody had pressed Fill the todo! that day. Claude returned each loop with its queue in order and the task it was on, and offered to run the arming pass if we asked. It didn't run it.
This step is worth doing even if you plan to give full access. It shows you what the assistant sees, and how it names your loops and tasks, before it touches anything.
Step 3: fill the todo, once, when you ask
Fill my DoLoop todo for today, then tell me what's on it.This is the same action as the Fill the todo! button: each loop arms its next task, and scheduled tasks due today are armed too. In our test, Claude first checked that today's list was still empty, ran the arming pass once, and listed six tasks: two scheduled ones and one from each of four loops.
DoLoop never arms anything on its own, and the MCP server keeps that rule. The server tells the assistant to run the arming pass only when you ask for it. It is also marked as a destructive tool, because running it twice in one day is not harmless: the second run re-arms tasks you have checked or skipped since and moves loops on. Ask for it once a day, the way you would press the button.
Step 4: check or skip a task by name
I've cleaned the sink. Check it off, and tell me what that does to the Kitchen loop.Claude found the armed task called "Sink", checked it, and its counter went from 7 to 8. It also explained the part people tend to miss: the Kitchen loop doesn't move on when you check a task. The next task comes up on your next arming pass.
Name the task that is armed now. When we asked Claude to check a kitchen task whose turn had already passed, it refused and said why: DoLoop only counts a check on an armed task, so the click would have changed nothing. If you aren't sure, ask "what's armed in Kitchen?" first.
Skipping works the same way: "skip the windows this round" uses the Abort tool. The skip is recorded, the loop advances on the next arming pass, and the task returns when the loop wraps. The assistant can't set the counters directly. It records progress only through check, abort and the arming pass.
Disconnecting
Each approval is a connection listed under My Profile → Connected apps, with its access level and when it was last used. Disconnect revokes it immediately. We disconnected our test connection and asked Claude Code for the list again. It reported that the doloop server needed authentication again and read nothing.
Changing or resetting your DoLoop password disconnects every app, and so does deleting your account.
Limits worth knowing
- 60 requests per minute per connection. A normal conversation stays far below that.
- Deleting a task or loop erases its counters for good. The assistant is told to confirm with you before it deletes anything, but read what it proposes before you agree.
- Only Claude Code was tested here. Claude Desktop and the web version of Claude, ChatGPT and Cursor can connect with the same URL. Their setup steps are on the developer page, but we haven't walked through those clients for this article.
- No secrets in chat. With the browser sign-in there is no key to paste. If you use an API key for a script instead, keep it out of prompts and out of git.
Where to go next
The developer page has the server URL, the setup for other clients and the API reference. If loops are new to you, start with why recurring tasks pile up. It explains arming, checking and skipping without an assistant in the way. DoLoop is available in English, French, German, Spanish, Italian and Chinese.