Skip to content
ProjectBrain Docs
Work in progress: these docs are being written and change often.

Connect an AI assistant

How Claude, ChatGPT and Codex connect to ProjectBrain, what they can do, and how to limit or revoke them.

GAOwnerExecutiveTeam leadIndividual contributorLast updated 5 October 2026

You can connect an AI assistant (Claude, ChatGPT or Codex) to ProjectBrain. Then you can ask it in plain words to list your tasks, comment, log time or draft a document, and it does that work in ProjectBrain for you.

The connection uses MCP (Model Context Protocol), an open standard many AI tools support.

Before you start

Section titled “Before you start”
  • You need the mcp.connect permission. Most roles have it.
  • Your admin may limit which assistants or which people can connect. If Connect does nothing useful, ask them.

Who the assistant acts as

Section titled “Who the assistant acts as”

Two accounts are involved. See principal and actor in the glossary.

  • You are the person who connected it. Your permissions decide what it may do.
  • The bot (Claude, ChatGPT or Codex) is the user name that appears on what it writes, with your name recorded as the person it acted for.

The assistant can only see and change what you can. If you lose access to a project, so does your assistant.

Scopes: what you let it do

Section titled “Scopes: what you let it do”

When you connect, you approve a set of scopes, such as tasks.read, tasks.write or time.write. Each scope is one kind of access. Your permissions still apply on top of a scope: tasks.write does not let the bot edit a task you could not edit yourself.

To change scopes later, open Settings > MCP Connections and choose Edit scopes on a connection.

  • Removing a scope takes effect immediately.
  • Adding a scope you did not approve before means signing in again from the assistant.
  • When ProjectBrain adds new scopes, the connection shows new permissions available. Choose Add new permissions to approve them.

What it cannot do

Section titled “What it cannot do”
  • Mark a task complete, unless your admin has set bots to self-drive. By default, Review required is on and a person finishes the task.
  • Publish a blueprint, approve its own work or merge code without a person’s say-so.
  • See pricing, unless you turn on Share pricing with this AI for that connection.
  • See teammates’ email addresses.

Limits

Section titled “Limits”

Each connection has a per-minute limit for reading and a lower one for writing, and your organization has a daily cap. If the assistant says it hit a limit, wait a minute. If it happens daily, ask your admin about the cap in Settings > MCP Policy.

Check that a connection is working

Section titled “Check that a connection is working”

Settings > MCP Connections lists your active connections. Each shows when it was last used, with a status badge:

Badge Meaning
Live Used in the last 24 hours.
Stale Not used for between one and seven days.
Dead Not used for more than seven days. ProjectBrain emails you when a connection reaches this point.
Never Connected but not used yet.

Manage or revoke a connection

Section titled “Manage or revoke a connection”

On each connection you can:

  • Rename it, so you can tell two connections apart.
  • IP Restrict it to the network ranges you list.
  • Set Expiry, so it ends by itself on a date within the next year.
  • Revoke it. You confirm with your password. The assistant loses access at once and must connect again to come back.

Guides for each assistant

Section titled “Guides for each assistant”

To have assistants pick up approved tasks on their own, see Agent orchestration.