How to Automate WhatsApp with AI in 2026 (No Code)

Somewhere in your WhatsApp there is a message you meant to answer on Tuesday. A client who wanted to move a call. A supplier who needed a number. Nothing tells you who is waiting, so the replies you owe sit there for days.

You can fix that with one sentence typed to an AI assistant. This guide shows the one recipe I would start with, how to connect it in four steps, and the shorter recipes you can add once it works. Every step names the limit it runs into, because automations that ignore their limits are the ones that break. If you want scheduled and recurring sends instead of drafts, see how to automate WhatsApp messages.

What does it mean to automate WhatsApp with AI, and what do you need?

Automating WhatsApp with AI means an assistant reads your chats and drafts replies or schedules messages from plain-English instructions. You need your WhatsApp number linked in Blueticks and the Blueticks connector added to Claude. The connector is included in every paid tier, and there is also a free plan with no time limit.

MCP, the Model Context Protocol, is an open-source standard for connecting AI applications to external systems. The MCP documentation describes it as being like "a USB-C port for AI applications." Blueticks runs a hosted MCP server, so Claude can reach your chats once you authorize it. This is the same idea behind most "claude whatsapp integration" and "whatsapp mcp" setups. It is not the first WhatsApp MCP, and it does not need to be.

Two things to be clear about up front.

• It runs on your own number. Blueticks works over WhatsApp Web through its own engine. It is not the official WhatsApp Cloud API, so there is no Meta business portfolio to set up.
• It acts when you ask. It is an ai whatsapp assistant you direct, not a bot that decides to message people on its own.

The connector is listed as working with Claude and ChatGPT on the Blueticks homepage and on the pricing page. For a longer walkthrough of what the tools can do, read the WhatsApp MCP guide.

How do you connect WhatsApp to Claude in four steps?

Link your WhatsApp number in Blueticks first, then add the hosted connector inside Claude. Open Settings, then Connectors, then Add custom connector, paste the Blueticks URL, press Connect, and approve on the Blueticks consent screen. That is the whole setup. There is no code to write.

The Claude integration docs say it plainly: make sure WhatsApp is connected in Blueticks first. Here are the four steps.

1. Link your number. Create an account at Blueticks, then link your WhatsApp number to it.
2. Add the connector in Claude. Go to Settings, then Connectors, then Add custom connector. Give it a name and paste https://api.blueticks.co/mcp.
3. Press Connect. Claude sends you to the Blueticks consent screen. The docs describe the sign-in as OAuth 2.0 with PKCE.
4. Authorize. Approve on the consent screen. For the recipe below, choose read-only access if the screen offers it. I explain why in the safety part of the recipe.

If you work with a developer, there is a local alternative. The @blueticks/mcp package reads a BLUETICKS_API_KEY from its environment and works with Claude Code, Claude Desktop and Cursor, per the MCP docs. Most people should use the hosted connector.

Quick check that the link is live. The API's GET /v1/ping call returns a whatsappConnections list, and an empty list means nothing is connected yet, as the integrations docs explain. If Claude later returns nothing for your chats, check the connection before you debug anything else.

Which chats are waiting on your reply right now?

Ask Claude for the chats where the last message is theirs, sorted by how long each has waited, then ask it to draft a reply for the top three. Tell it to send nothing. You get a ranked list and three drafts in one answer. You paste each draft into WhatsApp and send it yourself.

This is the recipe I would run first, because the result is visible in minutes and the risk is low. It reads, it ranks, it drafts. Nothing leaves your phone until you send it.

Write it as a single message. When the whole job is in one prompt, Claude can plan its reads together instead of asking you for the next step. Here is the prompt:

"List the WhatsApp chats where the last message is from them, not me, and tell me how long each has been waiting, longest first. Read the latest few messages in the top three and write a short reply draft for each, in my tone. Do not send anything. I will send them myself."

Notice what is not in the prompt. There is no date range to get wrong, no group filter, and no instruction to catch you up on everything. The question is narrow: whose turn is it? A narrow question gets a short, checkable answer.

Under the hood, Claude uses the chats_read tool. The MCP docs list its actions as list, search and list_messages; pulling older messages from the phone is a separate load_more_history action in chats_write. You never type those names. You type the intent, and Claude picks the action.

What does it look like on three example chats?

Here is an illustration on three invented chats. The names and details are made up, and your own list will look different. The shape of the answer is what matters: a ranking first, then one draft per chat.

Claude returns something like this:

1. Supplier, waiting 4 days. They asked for your purchase order number so they can confirm the shipment.
2. Client, waiting 2 days. They asked whether you can move Thursday's call to Friday afternoon.
3. Friend, waiting 1 day. They asked whether you are free for dinner on the 14th.

Then the drafts:

Supplier: "Sorry for the delay. The PO number is below. Can you confirm the ship date once it is in your system?"

Client: "Friday afternoon works for me. Does 3pm suit you? If so I will send an updated invite."

Friend: "Yes, I am in for the 14th. Where were you thinking? I can book if you tell me the time."

You read each one, change what is wrong, and paste it into the right chat. The supplier draft above says "below" because Claude cannot know your PO number. That is the kind of gap you fix by hand before sending. If a draft references something it could not have known, it is a placeholder, not a fact.

The numbered list is the part worth keeping. Four days is a different problem from one day, and seeing the wait time stops you answering the loudest chat first.

How many of your free calls does this use?

On the free plan, the API and the MCP share one quota of 5 requests per 6-hour window. Each MCP tool call counts once, and calls return a 429 until the next window. Claude decides how many calls it makes for this prompt; the docs publish no count, so plan for a few. Drafting happens in Claude, so drafts cost no calls.

The numbers come from the integrations docs, which also say any paid plan is unlimited. I will not promise you an exact count. Claude chooses its own reads, and a chat with a long history can take an extra one. The quota is the only hard number here; the call count is Claude's choice.

What this means in practice: your first run fits inside the free window, but a second prompt right after it may not. If you hit a 429, you wait for the next window or switch plans. The pricing page lists a free Pro trial with no credit card, which removes the cap while it lasts.

Ready to try it? Create a free Blueticks account, link your number, add the connector, and ask who is waiting on your reply. The first list and three drafts fit inside the free plan.

How do you make sure Claude only drafts and never sends?

Use two layers. In the prompt, say "draft only, do not send." In the connection, authorize read-only access, because the Claude docs state that when the connection was granted read-only access, write actions, including sending, are not offered. The second layer does not depend on Claude following your wording.

Prompt discipline is a request. Read-only access is a restriction. For this recipe you want both, and if you must pick one, pick the restriction. A read-only connection cannot send even if a prompt goes wrong, because the send action is not offered to Claude at all.

I am not claiming Claude pauses to ask before it sends. The Claude docs do not say that. The ChatGPT page does, which I cover below. For Claude, your safety is the read-only choice plus your own eyes on every draft.

If you later want Claude to send or schedule, re-authorize with write access for that job and then go back. Keep the habit of reviewing every draft. The human-in-the-loop guide goes deeper on the approval pattern.

What breaks: the list is only as good as the history Claude can read. If a chat's older messages are not on your phone, the wait time can look shorter than it is. The load_more_history option on list_messages exists to pull older messages from the phone, so you can ask Claude to look further back on a chat that seems wrong. Also, a chat where you replied from another device may still show up, so skim the top three before you trust the order.

What else can Claude run on your WhatsApp once it is connected?

Once the connector works, the same setup handles four more jobs: a follow-up sequence for leads, a digest of busy groups, a controlled campaign, and natural-language reminders. Each is a short pointer here, with a longer guide behind it. Start with one, and only add the next after you trust the first.

Follow up leads from one sentence. Paste a list of numbers and describe the sequence. Claude can build an audience, draft the first message and run it as a campaign you review. A few days later, ask who has not replied and nudge only those people. The failure mode is resending to the whole list, which pings people who already answered. Always read for replies first. See the Claude connection guide for the connector side.

Digest the groups you never read. Ask Claude which group chats need you and why, and tell it to send nothing. It is a read-only job, so a wrong ranking costs you a few minutes of rereading. The full recipe lives in the WhatsApp MCP guide.

Run a plain-English campaign and keep control. A campaign is a bulk send built from an audience and a message. You can ask Claude to pause, resume or cancel it. Use it only for people who expect to hear from you. WhatsApp's Terms of Service list "impermissible communications such as bulk messaging, auto-messaging" among the things you must not do. Consent and opt-out are the sender's responsibility. A campaign aimed at strangers is the fastest way to get flagged, so start with a small batch.

Schedule reminders in natural language. Tell Claude what to send and when, review the draft, and approve it. Check the date, time and time zone before you say yes, and check the chat after the send time instead of assuming it went out. The scheduling guide covers the details.

How is this different from a quiet-chat sweep or a weekly catch-up?

This page covers chats where the ball is in your court: the last message is theirs, and you owe a reply. A quiet-chat sweep looks at threads that went silent after you wrote. A weekly catch-up summarizes what happened. Different question, different list, different guide.

Picking the right one saves you a wasted run:

• Who is waiting on me? That is this guide. You get chats ranked by how long they have waited, plus a draft for each of the top three.
• Which threads went quiet? Read the Claude connection guide. It covers the quiet-chat sweep and a pre-call thread summary.
• What did I miss this week? Read the personal assistant guide. It covers the week catch-up summary.

Three more neighbours, in one line each. The chief-of-staff guide covers a wider assistant role. The inbox management guide covers sorting a crowded inbox. The group action items guide covers pulling tasks out of group chats.

If you use ChatGPT, the same server works there as a custom connector in developer mode. The integrations docs say reads and writes are separate tools, and ChatGPT asks before anything is sent or changed. The ChatGPT setup guide has the steps.

Where does the automation run, and when does it stop?

It depends on your plan. On Basic and Standard, Blueticks runs in your browser, so closing WhatsApp Web or turning off your computer stops it. On Pro, your agent, schedules and campaigns are cloud-hosted and keep working even when your computer is off. Compare plans on the pricing page.

Those are the pricing page's own words. The pricing page says: "On Basic and Standard, Blueticks runs in your browser." It also says: "If you close WhatsApp Web or turn off your computer, it stops working." And for Pro: "On Pro, your agent, schedules and campaigns are cloud-hosted and keep working even when your computer is off."

What that means for the lead recipe: it is a one-off question you ask while you are at your computer, so browser mode is fine. It matters more for the things that run later, such as scheduled reminders and campaigns. If a scheduled send must go out while your laptop is closed, you need the cloud plan. I will not quote prices here, since the page shows more than one figure per plan. Read it before you choose.

What breaks: on a browser plan, a closed laptop means a missed send. Check a scheduled message after its send time, every time, until you trust your setup.

How do you keep AI WhatsApp automation safe?

Keep the automation to what a careful person would do by hand. Read freely, message people who expect to hear from you, use campaigns for bulk sends, and review every draft. WhatsApp's Terms restrict automated and bulk messaging, so no unofficial tool can promise zero ban risk, and I will not either.

A few rules I keep for myself.

• The rules still apply to you. The connector sends as you, from your own number. WhatsApp's Terms of Service restrict bulk messaging and auto-messaging, and those rules cover what Claude sends on your behalf.
• Consent is your job. Blueticks cannot know whether someone wants your message. Who you contact, and how a person opts out, is the sender's responsibility.
• This is not the official API. Blueticks runs over WhatsApp Web, not Meta's Cloud API. The Cloud API documentation describes the official route, which is a different product with different rules.
• Linked devices can unpair. If you remove the linked device from your phone, the connection ends. Link it again to resume.
• A human reviews every draft. This is the point of the lead recipe. You are the last step before anything reaches a person.

For the approval pattern in detail, see the human-in-the-loop guide.

FAQ

Can I automate WhatsApp with AI without writing code? Yes. Link your number in Blueticks, add the hosted connector in Claude's settings, authorize it, and describe what you want in plain English. Developers can use the local @blueticks/mcp package instead, but none of the recipes here need it.

Does it use my own WhatsApp number? Yes. Blueticks links your existing number and works over WhatsApp Web through its own engine. It is not the Meta Cloud API. On Basic and Standard it runs in your browser, and on Pro it is cloud-hosted.

Will automating WhatsApp with AI get my number banned? No tool can promise that. The lead recipe only reads and drafts, which is the lowest-risk use. For sending, message people who expect to hear from you, start small, and read WhatsApp's Terms of Service, because those rules apply to you.

Do I need Meta Business verification? Not for this route, which uses your own number over WhatsApp Web. The Cloud API is a different route with its own requirements.

Does Claude send messages on its own? No. It acts when you ask, and for the lead recipe you tell it to draft only. If the connection was granted read-only access, write actions, including sending, are not offered, according to the Claude docs.

What happens when I hit the free quota of 5 calls? Further calls return a 429 until the next 6-hour window starts. Paid plans are unlimited, and the pricing page lists a free Pro trial with no credit card. Drafting inside Claude does not use calls.

Start automating your WhatsApp with AI today

Connect WhatsApp to Claude and ask one question: who is waiting on my reply? Link your number in Blueticks, then in Claude open Settings, Connectors, Add custom connector, and paste https://api.blueticks.co/mcp. Your first list and three drafts fit inside the free plan, and the free Pro trial lifts the cap with no credit card.

Authorize read-only, send the prompt from the lead recipe, read the drafts, and send the good ones yourself. Ten minutes from now, the reply you owe since Tuesday is gone. Create your account and start with that one.