Connect to PacedMind Cloud

With a PacedMind Cloud account, Claude Code, Codex and other MCP clients use PacedMind Cloud's MCP server. It serves your account's planner from any computer and any folder, whether or not the desktop app is running. No token goes into an agent's settings: each agent signs in with your account in the browser, and you allow it on PacedMind's page.

Sessions that PacedMind starts in a terminal don't use it: they reach the desktop app's own server on their computer, with a token that works only for their task. See MCP server.

pacedmind.com/connect has the steps for each agent: the commands to copy for Claude Code and Codex, Add to Cursor and Add to VS Code, and the address for Claude, ChatGPT and other clients.

Let your agent connect it

Paste this into Claude Code, Codex or another agent with a terminal:

Connect PacedMind's MCP server: follow "Instructions for agents" in https://pacedmind.com/docs/mcp/connect-cloud.mdx

The agent adds the server and starts the sign-in; the browser then opens PacedMind's page, where you select Allow.

With Connect in the desktop app

Signed in to PacedMind Cloud in the desktop app:

• After you sign in, PacedMind offers once, bottom right: Connect your agents to PacedMind Cloud. Connect sets up Claude Code and Codex, the ones installed on this computer; Not now leaves them as they are.
• Any time, in Settings → Connect your agents, select Connect next to Claude Code or Codex and confirm.

PacedMind sets the agent up for all your projects under the name pacedmind, in place of an earlier entry under that name or under organizer, and opens a terminal where the agent signs in (claude mcp login pacedmind or codex mcp login pacedmind). The agent opens a browser page: allow it there (see Allow the agent). Codex's earlier config.toml is kept as config.toml.pacedmind-backup.

Settings then shows the agent as Uses PacedMind Cloud, and Computers as PacedMind Cloud.

In the web app

Until an agent is connected to your account, the web app shows a card at the bottom right, Connect your agent, with a tab for each agent: Claude Code and Codex with their commands and Copy, Cursor and VS Code with Add to Cursor and Add to VS Code, and Claude, ChatGPT and Other with the address. The browser is signed in already, so the agent's sign-in only asks you to select Allow. Not now hides the card in that browser; Settings → MCP server has the same commands.

By hand

Claude Code

claude mcp add --transport http --scope user pacedmind https://app.pacedmind.com/api/mcp
claude mcp login pacedmind

The first command registers the server for all your projects; the second opens the browser to sign in. Inside Claude Code, /mcp also offers Authenticate for a server that needs it. The Claude app's Code sessions use the same registration.

Codex

codex mcp add pacedmind --url https://app.pacedmind.com/api/mcp
codex mcp login pacedmind

This adds the section below to ~/.codex/config.toml, which the Codex command-line tool and the Codex app share, and signs in:

[mcp_servers.pacedmind]
url = "https://app.pacedmind.com/api/mcp"

Cursor and VS Code

Add to Cursor and Add to VS Code on pacedmind.com/connect open the editor with PacedMind's server filled in: select Install there. By hand, add it to the editor's mcp.json.

Cursor (~/.cursor/mcp.json, or .cursor/mcp.json in a project):

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

VS Code (mcp.json in your user settings, or .vscode/mcp.json in a workspace):

{ "servers": { "pacedmind": { "type": "http", "url": "https://app.pacedmind.com/api/mcp" } } }

When the editor starts the server, it asks you to sign in, and the browser opens PacedMind's page (see Allow the agent).

Claude

Open Claude connectors, then Customize → Connectors → + → Add custom connector. Name it PacedMind and use https://app.pacedmind.com/api/mcp. Select Add, then Connect, sign in to PacedMind and select Allow. On Team and Enterprise, an owner first adds the connector for the organization.

Enable PacedMind in the conversation's connectors menu. The remote connection uses your Cloud account and works without the PacedMind desktop app running.

ChatGPT

In ChatGPT, turn on Developer mode under Settings → Security and login, then open Plugins, select +, name it PacedMind and enter https://app.pacedmind.com/api/mcp with OAuth. Let ChatGPT register its OAuth client; no token or client secret needs copying from PacedMind. Sign in on PacedMind's page and select Allow.

Start a new chat and add PacedMind from the tools menu. Developer mode availability depends on your account and workspace policy. These are the custom-connection steps while PacedMind's public directory listing is being prepared; they do not mean the listing has been approved.

Try: “What should I work on today?”, “Add a task to prepare the launch checklist tomorrow”, or “Move tomorrow's unfinished tasks to Friday.” Check the dates and proposed changes before approving them.

If connecting fails, confirm that the address is the public Cloud endpoint above, reconnect from the agent's settings, and sign in to the same PacedMind account. With two-factor sign-in enabled, finish its code step. OpenAI's connection guide and Claude's connector guide cover account and administrator restrictions.

Other MCP clients

Any MCP client that supports OAuth sign-in: add https://app.pacedmind.com/api/mcp as the server's address and follow its sign-in. The client finds PacedMind's sign-in on its own (OAuth protected resource metadata at https://app.pacedmind.com/.well-known/oauth-protected-resource/api/mcp) and registers itself.

Settings → Connect your agents shows the same commands with Copy commands.

Allow the agent

The sign-in opens https://app.pacedmind.com/oauth/consent. If the browser isn't signed in to PacedMind, it asks you to sign in first, then comes back to the page. An account with two-factor sign-in enabled must also enter its code. Without an account, Create an account there makes one: once you confirm your email in the same browser, it comes back to the page too. You can set up your authenticator later in Settings → Account. Planning and connecting MCP work before that; controlling computers requires two-factor sign-in and a new agent approval after setup.

The page asks Allow Claude Code to use PacedMind?, with the agent's name as the agent gives it. It shows the server the agent connects to, app.pacedmind.com (PacedMind Cloud), and where the sign-in is handed back to the agent: for Claude Code and Codex, an address such as 127.0.0.1:61154 or localhost:53963 on this computer, where the agent waits only while it signs in. That address isn't the MCP server. An agent you allowed before asks Allow this agent to use PacedMind again?

• Allow finishes a local agent's sign-in in a temporary window. Once the connection is confirmed, that window closes and PacedMind shows Connected to PacedMind. Return to your agent to continue; you can close the PacedMind tab or select Open PacedMind. Hosted clients, and browsers that block the temporary window, return to the agent's own page instead.
• Deny tells the agent no. For an agent you allowed before, the page says Not connected. You can close this tab.

If PacedMind shows Connection not confirmed, check the agent's sign-in window and select Check connection once it finishes. If the sign-in has ended, start connecting the agent again.

Only allow a request you started yourself, from an agent that returns somewhere you recognize. A request that waited too long says This request has expired: start connecting the agent again.

Instructions for agents

For an AI agent asked to connect PacedMind. Run the commands in the user's terminal; never ask the user for a password, code or token.

1. Check whether PacedMind is set up already: claude mcp list (Claude Code) or codex mcp list (Codex). An entry pacedmind with the address https://app.pacedmind.com/api/mcp that shows as connected needs nothing more; stop and say so.
2. Remove an older entry that points elsewhere, such as http://127.0.0.1:4319/api/mcp or the name organizer, only if the user agrees: claude mcp remove pacedmind --scope user, claude mcp remove organizer --scope user; for Codex, codex mcp remove pacedmind.
3. Add the server:
   • Claude Code: claude mcp add --transport http --scope user pacedmind https://app.pacedmind.com/api/mcp
   • Codex: codex mcp add pacedmind --url https://app.pacedmind.com/api/mcp
4. Sign in: claude mcp login pacedmind or codex mcp login pacedmind. It opens the browser. Tell the user to sign in to PacedMind if asked and to select Allow on the page Allow … to use PacedMind?. Wait for the command to finish.
5. Check: claude mcp list or codex mcp list shows pacedmind as connected. The tools are available in new sessions: in Claude Code as mcp__pacedmind__… (for example mcp__pacedmind__get_overview); a running Claude Code session picks them up after /mcp.
6. Use the server name pacedmind: PacedMind's skills and the sessions it starts refer to it by that name.

Other clients: add https://app.pacedmind.com/api/mcp as a streamable HTTP MCP server with OAuth, and let the user allow it.

What agents can do on PacedMind Cloud

An agent you allowed acts as you: it reads and changes areas, projects, tasks, the calendar and your preferences, sees your computers (list_computers), and reports on agent sessions with start_task, report_progress and finish_task.

Once you set up two-factor sign-in and reconnect the agent with it, two things it can only ask one of your computers for:

• A new session, with start_session: it goes to the computer the agent names, else the task's, else the default one. An agent never counts as having a two-factor code, so only a computer whose Two-factor code is off takes the request, and then that computer's From elsewhere setting decides: Refuse, Ask me (it waits for you there) or Start (at once). Where the code is on, the agent gives you a link to the task instead, where you start it with Start and your two-factor code, or you start it in the desktop app. See start_session.
• A folder for a project, an area or a task, with set_folder: the request waits up to a day for you to allow it in PacedMind on that computer, which checks the folder first. See set_folder.

It can't:

• resume sessions or send them back with changes: request_changes answers with a link to the task, where you do it with Request changes and your two-factor code, or in the desktop app;
• ask questions that wait for your answer in PacedMind: there is no ask_user, so it asks in its own conversation, and report_progress with kind question shows the question on the task;
• attach images from files on its computer: there is no attach_image, and finish_task takes no images;
• change your computers or their settings, answer for you, delete your account, or see the account's notification keys.

Everything else is the same as on the desktop app's server, with 40 of the 42 tools.

Connected agents

Settings → Connected agents lists the agents you allowed: each with its name and Signed in with the date, or Allowed, waiting for it to sign in until its sign-in is confirmed. Disconnect ends every sign-in of that agent at once, after the confirmation Disconnect Claude Code? It loses access to PacedMind at once.; to use PacedMind again, it signs in and you allow it again.

Signing out everywhere (Settings → Account → Sign out all) also ends every agent's sign-in.

Security

• An agent's sign-in counts only once you allowed it on PacedMind's page. Without an authenticator, that approval permits planning; it cannot control computers. When two-factor sign-in is enabled, finish it before allowing an agent. Each approval works for the one sign-in that comes of the request you allowed.
• If another sign-in of the same agent starts in the same few minutes, as when someone who has your password tries to slip in, neither counts: both are signed out, and connecting the agent again asks you again. If that happens without a reason you know of, change your password.
• The database itself checks every request against that approval, so a sign-in you didn't allow reads nothing, and one you disconnected stops working at once, not when its token expires.
• An agent's sign-in never counts as a two-factor session, even if it verified a code itself, so it can't do what needs a fresh code: resuming sessions, answering for you, or starting a session on a computer that asks for a code.
• Switching a computer's two-factor code off lets your agents ask it too. An agent you connected to PacedMind Cloud, or one on another of your computers, can then ask that computer for sessions, and with Start they start at once, in a folder of that computer's, on a task whose text the agent may have written. Keep the code on for any computer where you wouldn't let an agent start another agent without you. See General.
• A folder an agent asks for is only a suggestion: the computer checks that it's there, shows it to you, and sets it only when you allow it. Read the path before you allow it: sessions for that project, area or task start there.
• Agents keep their sign-in in their own credential store; nothing about it goes into their settings files.
• Requests from web pages (with an Origin header) are refused.

PacedMind's security overview covers the rest.