Claude and ChatGPT

Connect WORK³ to Claude or ChatGPT and work with your projects and items straight from the chat: “What is due for me this week?”, “Summarise where the LCH project stands”, “Create an item for the launch post and give it to Bo”. WORK³ is an MCP server, the standard way AI assistants connect to other tools, so other assistants that support MCP with OAuth work the same way.

What the assistant can do

• See your projects, items and comments, search them and summarise them.
• Create items, change them (status, assignee, due date, priority, labels and more), archive or restore them, and add comments, if you allow it.
• Create projects (from a template if you like) and change them: name, description, lead, client, value, contact and dates, or archive them, if you allow it and your role may manage projects.
• Make sub-items, and link items as dependencies (“Design blocks Build”) or remove such links.
• Read your chats with people and your projects’ chats, and send messages in them, if you allow it. It does not mark anything as read.
• List the files on items and projects, read text files (.txt, .md, .csv) and images, and save notes or summaries as text files.
• See a project’s boards and what is in each bucket, move cards between buckets, and add boards and buckets.
• Read your workspace’s workflows, create new ones as drafts and change drafts, if you allow it and your role may build workflows. It never publishes, turns on or runs a workflow: you check the draft in the editor and publish it yourself. It gives you a link to the draft each time.

It cannot change settings, statuses, members or roles, and it cannot delete anything (apart from removing a link between two items).

It acts as you: it sees only what you can see in WORK³, it can do only what your role allows, and what it does shows under your name, marked with the assistant’s name. It never sees other workspaces than the one you pick.

Connect Claude

1. In WORK³, go to Settings → Integrations → Claude and copy the server URL.
2. In Claude, open Settings → Connectors and choose Add custom connector. On Team and Enterprise plans an owner may need to add it for the organization first.
3. Name it WORK³, paste the server URL and leave the advanced settings empty.
4. Choose Connect. Claude opens WORK³: sign in, pick the workspace and choose Allow.

Connect ChatGPT

1. In WORK³, go to Settings → Integrations → ChatGPT and copy the server URL.
2. In ChatGPT, open Settings → Plugins and turn on Developer mode in the menu at the top right. It works on Plus and higher plans. On Business and Enterprise plans an admin may need to allow it.
3. Choose Create, name it WORK³, paste the server URL and pick OAuth as the authentication.
4. Sign in to WORK³, pick the workspace and choose Allow. In a chat, turn WORK³ on under the + menu.

Connect other MCP apps

WORK³ is an MCP server, so other assistants and editors that support remote MCP servers with sign-in can connect too, such as Claude Code, Cursor or VS Code.

1. In WORK³, go to Settings → Integrations → Other AI assistants and copy the server URL.
2. Add it as a remote MCP server in the app (Streamable HTTP). No API key is needed.
3. When the app asks you to sign in, sign in to WORK³, pick the workspace and choose Allow. Check that the address on the page is the app you expect.

The approval page

When you connect, WORK³ shows who is asking and where your approval goes. For Claude and ChatGPT it says Verified, because the answer can only go to their own sites (claude.ai, chatgpt.com). For any other app it shows the address in a warning: only continue if you just added WORK³ in that app yourself.

You choose:

• Workspace: the one the assistant works in. To use it in several, connect once per workspace.
• Create and change items and projects, and add comments: turn it off and the assistant can only read.

See what an assistant did

Everything an assistant does shows under the name of the person who connected it, with a small via Claude (or via ChatGPT) mark next to the name: in an item’s activity and comments, in the feeds on Home and the project page, in chats, in the Inbox and in the audit log. Other apps are marked with their name and address, for example via Cursor (localhost), so an app that only calls itself Claude cannot pass as it.

Allow or turn off assistants for the workspace

Under Settings → Integrations → Rules, people who manage integrations choose who may connect AI assistants (off, chosen roles or everyone; by default everyone but guests) and which ones: All MCP assistants or Only Claude and ChatGPT. It applies at once, also to connections that already exist.

See and revoke connections

Settings → Integrations → Claude (or ChatGPT, or Other AI assistants) lists your connected assistants with the assistant’s logo, a badge for what it may do (Read and change or Read only), who connected it, and when it was connected and last used. Revoke stops one at once. People who manage integrations see everyone’s connections in the workspace and can revoke them too. A connection also stops if you leave the workspace.

Security

• Sign-in uses OAuth 2.1 with PKCE, the current standard. The assistant never sees your password or your WORK³ session.
• Access tokens last an hour and are renewed with a refresh token that changes every time. If an old one is ever used again, WORK³ assumes it was stolen and ends the connection.
• WORK³ keeps only fingerprints (hashes) of codes and tokens, never the tokens themselves.
• Every connection and revocation is in the audit log.
• WORK³ sends data to the assistant only when you ask it something. What Anthropic or OpenAI does with it is governed by your own agreement with them; see Services you connect.