Many of the people who use Outreach Asia already spend part of their day in Claude or ChatGPT. Until now, they had to copy a draft or a reply out of the app to ask an assistant about it. From today, an assistant can read the workspace directly, and with your permission it can help tidy the review queue.
Your assistant. Your workspace.
- Choose access
Start with a read-only token, or grant read-and-act access.
- Connect
Use the workspace connection in your assistant.
- Review actions
Edits are checked and begin a fresh 24-hour review.
A visual summary of this article. Read the full explanation below.
Why we built it.
The questions clients ask us most are simple ones that take several clicks to answer. What sends today? Did anyone reply? Which campaign is working? An assistant connected to the workspace can answer them in one sentence, in English, Bahasa Melayu or Chinese, and it can combine them: for example, “review everything due in the next 24 hours and tell me which drafts misread the company”.
How it connects.
The connection uses the Model Context Protocol (MCP), an open standard that Claude, ChatGPT and many developer tools support. In Settings → AI assistants, a client creates a personal access token and chooses Read only or Read and act. They then paste a secret link into the assistant's connector settings, or use the token as a header in Claude Code and other developer tools.
Each token belongs to one workspace, is shown once and is stored only as a hash. A client can create one token per assistant and revoke any of them, and a revoked token stops working immediately.
What an assistant can do.
- Summarise the workspace: campaigns, today's research, drafts waiting for review and replies needing attention.
- List and explain leads, including why each company was chosen and the evidence from its own website.
- Read drafts with their statement-check results and when their review ends.
- Read replies and whole conversations, including a suggested answer waiting for approval.
- Report results by period and by campaign, counted as the Reports page counts them.
- With a Read and act token: skip a company, edit a draft that is waiting for review, or approve a draft revision.
What it cannot do.
An assistant has no way to send a message or to make one send sooner. When it edits a draft, the statement checks run again and a new 24-hour review window starts, exactly as when a person edits in the app. Approving a draft does not shorten the wait either. An assistant cannot contact anyone who has opted out, cannot change sending or mailbox settings, and cannot delete anything. Each change it makes is recorded in the workspace activity.
We set these limits on purpose. The rules that protect recipients and senders should not depend on who, or what, is at the keyboard.
How clients are using it.
The patterns we expect to matter most are small and daily ones:
- A morning check of what is due to send and what needs a decision.
- A second opinion on drafts before a batch goes out.
- A plain-language reading of what a buyer is actually asking for in a reply.
- A Friday comparison of campaigns.
Our help centre guide includes prompts for each of these that can be copied as they are.
Your data.
When a client connects an assistant, whatever the assistant reads from the workspace is sent to that assistant's provider and handled under the client's own agreement with them. We recommend starting with a Read only token, checking your organisation's policy first, and revoking a token as soon as it is no longer needed. The step-by-step guide covers setup for Claude, ChatGPT and Claude Code, the full list of tools and troubleshooting.
This article describes the product as released on 29 September 2026. Details may change as the product develops; your workspace always shows the controls available to you. The menus in Claude and ChatGPT are controlled by Anthropic and OpenAI and may change. Nothing here is a guarantee of replies, meetings or inbox placement.