Docs

Your agent

ScrollBuddy is an MCP server. Connect the agent you already talk to once, and it plans the day with you and checks the work.

Before you connect

Connecting opens a sign in: your phone number, a six digit code that comes on WhatsApp, then a card asking how hard your agent should check, which is the dial. Then Allow. Invite only for now. UK and EU numbers; the code comes on WhatsApp.

Each agent you connect has its own work switch, chosen when you connect it.

The iPhone app that signs in and works with your agent is in testing, and so is Android. The App Store has the earlier app, which locks apps on its own. To start now, get the earlier app on the App Store; the new app comes with your spot.

Claude Code

Run this once in any terminal.

claude mcp add --scope user --transport http scrollbuddy https://api.scrollbuddy.app/mcp

Then type /mcp in Claude Code, pick scrollbuddy and choose Authenticate. A browser opens for the sign in.

Claude

Open Customize, then Connectors. Choose Add, then Add custom connector. On a Team or Enterprise plan an owner adds it under Organization settings, then Connectors.

https://api.scrollbuddy.app/mcp

Name it ScrollBuddy, paste this URL, choose Continue, then Add. Sign in when it asks.

ChatGPT

In Settings, open Apps, then Advanced settings, and turn on Developer mode. Then choose Create app next to Advanced settings.

https://api.scrollbuddy.app/mcp

Name it ScrollBuddy, paste this URL, pick OAuth and sign in when it asks.

Cursor and every other host

Any host that speaks MCP over HTTP with OAuth works. Add a server with this URL.

https://api.scrollbuddy.app/mcp

In Cursor, add servers from Customize in the sidebar, or put one entry in ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{ "mcpServers": { "scrollbuddy": { "url": "https://api.scrollbuddy.app/mcp" } } }

House rules

Paste one paragraph into your agent’s own instructions: CLAUDE.md, AGENTS.md, Cursor rules, Claude project instructions, or a custom GPT’s instructions. Pick the one that matches your dial.

Easy

When I say a win is done, take my word for it and mark it done in ScrollBuddy. If you saw the work happen, say what you saw in one line. Never mark a win done that I did not tell you about, never name one I did not choose, and never spend my credits unless I ask. When a piece of work finishes and you can point at the result (a commit, a page, a decision written down), log it as a win; judge my part from what you saw in this conversation, and ask me only when the work happened where you could not see it. Credits follow my part: the full amount for what I did myself, about a third for what I directed and reviewed, nothing for what I never looked at.

Normal, the default

When I say a win is done, ask what I can show you: a link, a file, or the text I sent. Mark it done in ScrollBuddy with one line in your own words naming what you checked. If I have nothing to show, mark it only when I have said it is done in my own words, and say that is what you went on. Never spend my credits unless I ask. When a piece of work finishes and you can point at the result (a commit, a page, a decision written down), log it as a win; judge my part from what you saw in this conversation, and ask me only when the work happened where you could not see it. Credits follow my part: the full amount for what I did myself, about a third for what I directed and reviewed, nothing for what I never looked at.

Strict

When I say a win is done, ask for proof before you mark it done: a merged PR, a published page, a file you can open. Check it yourself and write one specific line naming what you saw. If I have nothing to show, do not mark it done; tell me what would count. My word alone is not enough. Never spend my credits unless I ask. When a piece of work finishes and you can point at the result (a commit, a page, a decision written down), log it as a win; judge my part from what you saw in this conversation, and ask me only when the work happened where you could not see it. Credits follow my part: the full amount for what I did myself, about a third for what I directed and reviewed, nothing for what I never looked at.

What your agent can do

  • Read today’s page, your balance, the month’s goal and past days.
  • Name wins you choose, with their credits and their proof, and drop one you no longer want.
  • Show the step you are on, with the small squares.
  • Mark a win done once it has seen the proof, or log one you finished without naming it.
  • Mark the key win, the one you say mattered most today.
  • Set the month’s goal in your words.
  • Spend your credits on screen time, only when you ask.

What it cannot do

  • Pay a win without the proof you set when you named it, or below your dial. It is refused in one sentence and nothing pays.
  • Spend more credits than you have.
  • Reach your files, your repository or the page behind a link through ScrollBuddy. We keep a link’s host, never the page.
  • See your win photos or, on iPhone, which apps you locked.

ScrollBuddy runs no model on your agent’s calls. Your agent is the one who checks; the server holds it to your bar.

Your part

Credits follow your part. The full amount for work you did with your own hands, about a third for work you directed, reviewed and accepted, nothing for work you never looked at.

Your agent working on its own is not a win. Credits pay for your effort, never for compute.

The dial

The dial is how hard your agent and your phone check before a win pays: Easy, Normal or Strict. Change it in the gear on your phone or on your dashboard. Each level, in the app’s words, is on proof.

Disconnecting an agent

Your dashboard lists every connected agent with a Revoke button. Sign out everywhere ends every phone, browser and agent at once. More on your data.