Connect Atwork to Claude, ChatGPT, Cursor — or any assistant that speaks MCP — and ask about your real tasks, sheets and dashboards without leaving the conversation.
Pick the one you use — it takes about a minute.
In Claude (web or desktop), open Settings → Connectors → Add custom connector, paste this address, then click Connect:
https://mcp.atwork.so/mcpAnything that means reading your workspace or making a small, well-defined change to it.
Your assistant looks things up and adds to your work with the same access you have — nothing more.
workspace.search
Read
member.read
Read
my-day
Prompt
summarize-sheet
Prompt
sheet.read
Read
sheet.aggregate
Read
task.read
Read
task.create
Write
item.create_humanized
Write
item.update_humanized
Write
dashboard.read
Read
page.read
Read
page.create
Write
page.edit
Write
comment.post
Write
calendar.read
Read
file.request_upload
Write
file.confirm_upload
Write
A connected assistant can never see more than you can. Every permission is carved out of the access you already have — nothing is extended.
Any assistant that supports the Model Context Protocol — including Claude (web, desktop and Claude Code), ChatGPT and Cursor. Add Atwork in its connectors settings and sign in.
No. The connection only answers your assistant's requests in the moment, under your account's permissions. Atwork doesn't send your workspace anywhere else or use it for training.
Nothing extra — the connection is included with Atwork and works today on every plan, against the same workspace your team already uses.
More on working with AI in Atwork
Atwork speaks MCP, so Claude can read your sheets and create tasks directly. What it can do, what it cannot, and when this beats using an in-app agent.
Build a custom AI agent in Atwork, scope what it can see, assign it tasks like a colleague, and @mention it in comments. A practical setup guide.
Scope, capabilities, and identity: how to reason about what an AI agent can read and change in your workspace before you roll one out to your team.