Connect Claude or ChatGPT to your workspace

What this is

The Model Context Protocol (MCP) is an open standard that lets AI assistants call tools in other services. Outreach Asia runs an MCP server for your workspace at:

https://app.outreach.asia/mcp

An assistant connects with a personal access token that you create in the app. The token belongs to one workspace, so the assistant only ever sees that workspace. You can create several tokens, for example one per assistant or per laptop, and revoke any of them at any time.

You don't need any technical knowledge to use it. You create a token, paste one link into your assistant, and then talk to it in plain language.

What an assistant can and cannot do

When you create a token you choose one of two access levels.

With either level, an assistant cannot:

• send a message, or make a message send sooner;
• shorten the 24-hour review window. Editing a draft starts a new 24-hour window, as it does in the app;
• get round the statement checks. An edited draft is checked against your approved statements again;
• contact anyone who has opted out, or remove anyone from the do-not-contact list;
• change sending, mailbox, automation or billing settings;
• delete anything.

1. Create a token

1. Sign in at app.outreach.asia and open Settings → AI assistants.
2. Under Create a token, give it a name you'll recognise later, for example Claude on my laptop.
3. Choose Read only or Read and act, then select Create token.
4. The page shows Your new token and a Secret server URL. Copy the one you need with Copy token or Copy secret URL.
5. Select Done. For your security the token is shown only once. If you lose it, revoke it and create another.

There are two ways to connect, and your assistant decides which one you use:

Treat the token and the secret URL like a password. Until you revoke it, anyone who has either one can use your workspace at the token's access level.

2a. Connect Claude (claude.ai and Claude Desktop)

Claude calls external MCP servers custom connectors. Anthropic decides which plans can add them and changes the menu names from time to time. If a step below looks different, look for Connectors in Claude's settings.

1. In Claude, open Settings → Connectors and choose to add a custom connector.
2. Name it Outreach Asia and paste your Secret server URL as the URL.
3. Leave the advanced or OAuth fields empty, because the secret URL already identifies your workspace.
4. Save it. Then, in a new chat, switch the connector on from the tools or connectors menu under the message box.
5. Test it by asking: “Use Outreach Asia to give me a summary of my workspace.”

Claude Desktop uses the same connectors as claude.ai, so a connector you add on the web normally appears in the desktop app too. Claude asks before it uses a tool that changes something. Read those prompts rather than approving them by habit.

2b. Connect ChatGPT

ChatGPT connects to custom MCP servers through developer mode for connectors (also called apps). Whether it's available depends on your ChatGPT plan and your workspace administrator, and OpenAI changes the menu names from time to time.

1. In ChatGPT, open Settings → Apps & Connectors (or Connectors) → Advanced settings, and switch on Developer mode.
2. Create a new connector. Name it Outreach Asia and paste your Secret server URL as the MCP server URL.
3. Set authentication to No authentication, confirm that you trust the server, and create it.
4. In a new chat, pick the connector from the tools menu (+, then Developer mode or More) and ask for a workspace summary.

On a business or enterprise workspace, your ChatGPT administrator may need to allow custom connectors first. Like Claude, ChatGPT asks you to confirm before it uses a tool that changes something.

2c. Claude Code and other MCP tools

Claude Code and most developer tools can send the token as a header, which keeps it out of the URL. With Claude Code, run:

claude mcp add --transport http outreach-asia https://app.outreach.asia/mcp --header "Authorization: Bearer oa_your_token"

Then type /mcp inside Claude Code to check that outreach-asia is connected. Other MCP clients need the same two things:

• the server URL https://app.outreach.asia/mcp, which uses the Streamable HTTP transport;
• the Authorization: Bearer header.

If you write your own scripts, the server speaks JSON-RPC 2.0 over HTTP POST, with initialize, tools/list and tools/call. Each token can make up to 60 calls a minute.

The twelve tools

Your assistant picks the right tool by itself, so you don't need to name one. Knowing what exists does help you ask better questions.

With a Read only token, the assistant doesn't even see the three act tools.

Prompts that work well

Assistants do best when you give them a clear job, a scope and the output you want. Copy these and adapt them.

Morning check

Using Outreach Asia, give me this morning's summary: drafts whose review ends in the next 12 hours, any flagged by the checks, and replies that need me. Keep it to a short list with company names and one line each.

Review the drafts before they send

List the drafts due in the next 24 hours. For each one, read the draft and the lead, and tell me if the email misreads the company, is too long, or asks for too much. Don't change anything. Just give me a table: company, verdict, reason.

Tighten one draft (Read and act)

Get the draft for [company]. Rewrite the body to under 90 words, keep every factual statement exactly as it is, and keep the question at the end. Show me your version first. Only save it with edit_draft after I say yes.

Understand a reply

Open the conversation with [company]. What are they actually asking for, what does the suggested answer say, and is anything missing from it? Suggest what I should send, but don't send anything.

Weekly results

Compare last week with this week for each campaign: companies found, emailed, replies and interested. Which campaign is doing best per email sent, and what should I change? Keep it to five bullet points.

Look for patterns

Look at the Priority leads in Malaysia that haven't been emailed yet. Group them by type of business and tell me which groups look most like the companies that replied to us.

In other languages

You can ask in Bahasa Melayu or Chinese, for example “用 Outreach Asia 总结一下今天需要我处理的回复。” The assistant answers in your language, while the drafts themselves stay in the buyer's language.

Daily and weekly routines

• Every morning (5 minutes): ask for the morning check. Open anything flagged in the app, or ask the assistant to explain the flag.
• Before a batch sends: ask the assistant to review the drafts due in the next 24 hours. Then skip the companies that don't fit and approve the rest, in the app or with a Read and act token.
• When a reply arrives: ask what the buyer wants and what the suggested answer leaves out. Send the answer from the app.
• Every Friday: ask for the weekly results by campaign and one change to try next week.

Good habits

• Ask for a proposal before a change. Have the assistant show you an edit or a list of companies to skip before it acts.
• Keep facts in the app, not in the chat. An assistant can't add new statements. If a draft needs a fact you haven't approved, add the source document in the app and approve the statement there.
• Remember that edits restart the review. If you edit a draft that was due in an hour, it moves to 24 hours later. That is intentional.
• Check the numbers against the app. Results use the same counts as the Reports page. If an assistant's summary disagrees with Reports, trust Reports and tell us.
• Use one token per assistant. That way you can revoke one without disconnecting the others, and Last used tells you which ones are active.

Privacy and revoking access

When you connect an assistant, whatever it reads from your workspace goes to that assistant's provider, Anthropic or OpenAI, and is handled under your agreement with them. That includes company names, published business addresses, drafts and replies. Check your organisation's policy before you connect, especially for replies that contain a person's name or details.

Every change an assistant makes is recorded in your workspace activity, just as your own changes are.

To revoke a token:

1. Open Settings → AI assistants and find the token in the list.
2. Select Revoke, then Revoke now.

The token stops working immediately, and so does its secret URL. Revoke a token straight away if a laptop is lost, someone leaves your team, or a secret URL has been pasted somewhere it shouldn't be.

This article explains how Outreach Asia handles data you share with an assistant. It is not legal advice about your own obligations.

Troubleshooting