Claude WhatsApp Personal Assistant: Catch Up on a Week of Chats with One Prompt (2026)

Picture a four-day trip with the phone in a drawer. A Claude WhatsApp catch-up can get you back to speed fast. When you get back, the unread count is in the hundreds, and only three or four threads matter. This guide shows how to use Claude as a WhatsApp personal assistant, so you catch up on WhatsApp messages with one prompt and then approve the few that need you.

What does a WhatsApp personal assistant do when you come back after days offline?

A WhatsApp personal assistant AI reads the messages that arrived while you were away, groups them by thread, and shows you which ones are waiting on you. It drafts replies and holds them. It sends nothing until you say yes. With Claude, the reading runs through a hosted connector to your own WhatsApp number.

Two facts to know first. Blueticks runs your own number through a WhatsApp engine (a browser extension or a hosted gateway), not the Meta Cloud API. Meta's documentation says that a number already in use with WhatsApp cannot be registered unless it is deleted first, so the Cloud API is a separate product for this job. Second, Claude does not get pinged when a message lands. You ask it to check.

For the recurring version, a daily brief and triage rubric that work as a WhatsApp chief of staff AI, read the chief-of-staff routine. This article covers the one-off catch-up after you have been away.

How do you give Claude read access to your WhatsApp?

Add https://api.blueticks.co/mcp as a custom connector in Claude, then sign in to Blueticks to approve access. Claude stores no API key this way. Your number must be linked to an engine that is online, because reads fail with a 409 ENGINE_NOT_ONLINE error when it is not. For the click-by-click version, including the MCP-server path for Claude Code, see the connection guide.

1. Link your number at blueticks.co/signup and confirm the engine shows as connected.
2. In Claude, open Customize > Connectors. Team and Enterprise owners find it under Organization settings > Connectors.
3. On Pro or Max, click + Add, then Add custom connector. On Team or Enterprise, click Add, hover Custom, and select Web. Name it Blueticks, paste https://api.blueticks.co/mcp, and click Connect.
4. Sign in to Blueticks and approve the connection. This is OAuth, so your Blueticks password never goes to Claude.
5. Test with a read-only question, such as "List my WhatsApp chats with activity this week."

Check the plan limit before you start. Anthropic's connector help page says custom connectors over remote MCP are available on Claude, Cowork, and Claude Desktop, across the Free, Pro, Max, Team, and Enterprise plans. Free accounts get one custom connector. If you already use it for something else, this takes its slot.

What breaks: if Claude says it has no WhatsApp access, the connector is not signed in, or the engine is offline. Fix the connector in Customize > Connectors and check engine status in Blueticks before you test again.

Does it work from the Claude phone app?

Anthropic's help page says connectors work across every Claude client, including the mobile apps, and describes adding them on the web and desktop. We have not tested the Blueticks tools on a phone, so treat phone use as unverified. Add the connector on the web or desktop first, then check your phone app.

Here is the prompt for the catch-up:

"I was offline from Monday to Friday. Catch me up on my one-to-one WhatsApp chats since Monday 9am. For each thread where someone is waiting on me, give the name, how long they have waited, and one line of context. Skip group chats. Max 15 threads, longest wait first."

Claude works through it in order:

1. It calls the utils tool with current_date_time to turn "Monday 9am" into your local time. The connector instructions say to do this before any request with a time in it.
2. It lists chats with activity since that time, limited to one-to-one chats.
3. It reads the messages in the threads that look unanswered.
4. It returns the ranked list.

Claude only sees the history the engine has loaded (the WhatsApp MCP primer lists the read tools). Message reads (list_messages, search and load_more_history) return a coverage object that shows how far back the read actually went. If your window reaches past what is loaded, Claude can call load_more_history for one chat, up to 10 pages per call. Ask it to report the coverage it got. Do not trust "nothing from Tuesday" until the coverage says the read reached Tuesday.

Connect your own number at blueticks.co/signup, add the connector in Claude, and paste the catch-up prompt above. No Meta API account and no bridge to keep running on your laptop.

How do you get one summary that ranks what needs you first?

Ask for one list of threads where someone is waiting on you, with a time window, a cap, and one line of context per thread. The catch-up prompt already does this. A tight window and a hard cap rank the list far better than "summarize everything." The recurring triage rubric is in the chief-of-staff routine, so it is not repeated here.

A good result is short and sorted by who has waited longest (illustrative, not a real chat):

Waiting on you since Monday (3)

1. Supplier, 3 days: asked you to confirm the delivery slot.
2. New enquiry, 2 days: asked about pricing, no reply yet.
3. Client, 1 day: wants to move Wednesday's call.

Three levers do most of the work: a time window, a scope that excludes groups by default, and a cap. Without the cap you get timestamp order, which is the pile you were trying to escape.

What breaks: asking for too much. A two-week window across every group dilutes the ranking and runs into the coverage limit. If Claude says a thread had nothing relevant, check the coverage first. A truncated read means the answer is inconclusive, not "no."

How do you have Claude draft replies you approve before anything is sent?

Ask for drafts, tell Claude to show them to you, and approve each reply by name. The approval rules below are the ones that keep this safe. Claude sends a message only after you say so. Keep the Blueticks tools on per-call approval, so each send asks you first.

This rule is built into the protocol. The Model Context Protocol specification says that "there SHOULD always be a human in the loop with the ability to deny tool invocations" (MCP specification, tools). Anthropic's connector guidance is just as direct: only click "Allow always" when using a server and tool that you trust to run unsupervised.

Do not click "Allow always" on the Blueticks connector. Anthropic's wording applies "Allow always" to a server and tool, and the chats tool both reads and sends, so it can let a send run without asking.

1. Ask for drafts: "Draft a reply to each thread in the list. Two sentences maximum, no exclamation marks. Show them to me and do not send anything."
2. Correct weak drafts in one line: "Too formal. Drop the second sentence."
3. Approve one at a time: "Send the reply to the supplier." Claude uses the immediate chat send action, so the message leaves right away.

What breaks: a draft only matches the thread's tone as well as the thread shows it. Read each draft before you approve it. And if you clicked "Allow always" on the connector, step 3 can become a silent send.

How do you schedule the follow-ups you approved for later?

Ask Claude to schedule an approved message for a time you name. It converts the time, then queues the message with the scheduled-messages send action, passing the recipient, the text, and a send_at timestamp. Until it goes out, you can list, update, or cancel the message. The full timing rules are in scheduling WhatsApp messages from Claude.

1. Name the time in your own words: "Schedule the supplier reply for Wednesday at 8:30am, my time."
2. Claude calls utils with current_date_time, then works out the local timestamp.
3. Claude calls scheduled_messages with send, passing to (the recipient's chat id), text, and send_at.
4. Check it: "List my pending scheduled messages." The new message should appear with status pending.

What breaks: a timezone mistake. Claude should convert your local time to UTC with a trailing Z before it calls the tool, for example 2026-10-21T05:30:00Z. Also, once a scheduled message has been sent, the default pending list hides it, so ask for sent messages explicitly if you want to confirm delivery.

What should a personal assistant never do on your WhatsApp without you?

Never let the assistant send to a new contact, post in a group, or write about money or health without your explicit go-ahead. Put those rules in the catch-up prompt so they apply every time, not only when you remember to type them.

• New contacts. Only reply to people already in your chats.
• Groups. Summaries are fine. Posting is not.
• Money and health. Nothing about payments, invoices, or medical matters without you approving the exact wording.
• Read state. Do not mark chats read or archive them unless you asked. Both change what your own inbox shows.

These are rules you write into the prompt. Blueticks does not enforce them for you, so the approval step is the real gate. The assistant also ranks by the rules you gave it. It will confidently drop a thread into the noise pile for a reason you never wrote down, so check the skipped group chats every so often.

What can't Claude do with WhatsApp on its own, and how does the Blueticks connector add it?

Claude on its own has no access to your WhatsApp, so a Claude personal assistant for WhatsApp needs a connector to reach your messages. The Blueticks connector gives it a hosted MCP endpoint that reads and sends through a WhatsApp engine on your own number. It is not the Meta Cloud API, and no connector can promise your number will never be restricted.

Plan around three limits:

• Pull, not push. Nothing reaches Claude when a message arrives. You ask for the catch-up when you want it.
• The engine must be online. Reads fail until it is back, so check engine status before a trip.
• History is what the engine has loaded. Use the coverage object, and load older history for a specific chat when you need it.

On risk, WhatsApp's Terms of Service prohibit using the service for "sending illegal or impermissible communications such as bulk messaging, auto-messaging, auto-dialing, and the like" (WhatsApp Terms of Service). A catch-up is reading, which carries low exposure. Sending is where the exposure is.

Blueticks is a hosted, no-code connector. It is not the first WhatsApp MCP server. Open-source servers that run on your own machine came earlier, such as lharries/whatsapp-mcp. Our comparison of WhatsApp MCP servers covers the trade-offs.

Frequently asked questions

Can Claude catch up on WhatsApp without me opening the app? Yes, if the engine is online and the connector is signed in. It reads when you ask, and it does not message you when something arrives.

Will Claude send messages after I'm back without asking? No. Drafts are shown to you, and a send happens only after you approve it.

Can Claude read messages older than a few days? Only what the engine has loaded, plus what it pulls with the history action. That action goes up to 10 pages per call, and the coverage object shows how far back the read reached.

Is this the WhatsApp Business API? No. It runs on your own number through a WhatsApp engine (a browser extension or a hosted gateway). Cloud API numbers cannot be used with the regular WhatsApp app, so the two are separate products.

Will this get my number restricted? Nobody can promise that. Reading your own chats is low risk. Sending is where the exposure is, which is why every send should be one you approved.