Your WhatsApp is full of questions you have already answered. An AI agent can read them, draft the replies, and send the ones you approve. The catch is picking a route that still works after the first week, because the routes differ on who can run them, when they stay up, and what Meta allows.
This guide maps the four routes, shows the uptime each one really gets, and walks through building a Blueticks agent in plain words and testing it.
What is a WhatsApp AI agent, and which kind do you need?
A WhatsApp AI agent is an AI model connected to WhatsApp so it can read chats, draft replies, and send messages from plain-English instructions. Two different products share that name. A customer-facing bot sits on a business number and answers customers, such as order status or bookings. An operator agent works for you on your own account: it reads your inbox, drafts replies in your voice, and follows up on the leads that went quiet. Most solo operators want the second kind. The WhatsApp MCP guide covers that side in more depth.
Decide which one you are building before you compare routes. The two need different plumbing, and mixing them up is the costliest mistake in this space.
What are the 4 ways to put an AI agent on WhatsApp?
The four routes are Meta's official WhatsApp Business API, Claude connected to a WhatsApp MCP server, a Blueticks agent that you describe in plain words, and a no-code platform. They differ on Meta approval, where the agent runs, and whether it can act as you or only as a customer-facing bot.
- Official WhatsApp Business API: full control, Meta approval, template review
- Claude + a WhatsApp MCP server: no Business API onboarding, runs on your own number, needs a local process
- Blueticks agent: described in plain words, runs on your own number, uptime set by your plan
- No-code agent platforms: quickest start for a fixed-script support bot
Option 1: The official Business API
This is the route for customer-facing bots. You register a business number, pass business verification, and submit message templates for review. Meta's messaging limits start a newly created business portfolio at 250 unique customers in a rolling 24 hours. Reaching 2,000 takes business verification, or 2,000 delivered messages to unique numbers within 30 days using high-quality templates. Higher tiers unlock through automatic scaling after that.
Meta has also narrowed the door. TechCrunch reported that Meta's updated Business Solution terms, effective January 15, 2026, bar general-purpose AI assistants from the API, and that Meta said customer-service bots are unaffected. Scope can differ by region, so read Meta's current terms before you plan around it. Either way, a bot on the official API needs a defined business job, such as support, orders, or bookings.
What breaks: template review and the tier limits above can slow a launch. Budget time for both.
Option 2: Claude with a WhatsApp MCP server
MCP, the Model Context Protocol, lets an AI model call outside tools. A WhatsApp MCP server exposes WhatsApp actions, such as listing chats, reading a thread, and sending a message, as tools Claude can call. You register it once in your Claude config and Claude can work on your WhatsApp when you ask.
The cost is that the server lives on your machine. When the machine sleeps, the agent sleeps. This roundup of WhatsApp MCP servers compares the projects if you want to pick one.
Option 3: A Blueticks agent
You describe the agent in plain words, and Blueticks runs it on your WhatsApp account. The homepage puts it in one line: "Describe it in plain words. It answers your WhatsApp chats, you stay in control." Replies are instant, on your own number, in your customer's language. The agent knows when a human should step in and tells you.
Developers can also reach Blueticks through its API and MCP layer, which is included in every paid tier. The Blueticks API page publishes a rate limit of 60 messages per minute on Basic, 120 on Standard, and 300 on Pro, per phone number.
Option 4: No-code agent platforms
Flow-builder platforms fit a fixed-script support bot well. Judge them on the job you need, not the headline price, because seat fees and add-ons often push real spend past the sticker. For an operator agent that takes freeform instructions, check whether the platform does that at all before you sign.
Here is the same comparison in one table:
| Route | Meta approval | Where it runs | Best fit |
|---|---|---|---|
| Official API | Yes: verification and template review | Your server, on Meta's platform | Customer-facing bot at scale |
| Claude + self-hosted MCP | Varies by setup | Your machine | Solo developer, working on your own chats |
| Blueticks agent | Not required for the operator route on Basic and Standard | Your browser (Basic, Standard) or the cloud (Pro) | Operator agent on your own number |
| No-code platform | Varies by vendor | Vendor platform | Fixed-script support bot |
If you want an agent on the number you already use, create a Blueticks account and describe the agent in plain words.
How do you build a Blueticks agent in plain words?
Describe the agent, connect your number, decide how a human gets pulled in, then send a test message from a second phone. Start with one narrow job you can review. A first agent that does one thing well teaches you more than a broad one you cannot check.
-
Describe the agent. Write what it handles, in what tone, and what it should never promise. Plain sentences are enough.
-
Connect the number you already use. Open the Blueticks app and connect the WhatsApp account the agent will answer on. This is your own number, not a new business line.
-
Check how a human takes over. The agent knows when a human should step in and tells you. The public homepage and pricing page do not publish the exact rule, so do not assume one. Send a message that should reach a person and see what happens.
-
Send a test from a second phone. Message your connected number from another phone and check the reply. It should arrive on your number, instantly, in the language your customer wrote in.
-
Run it on a real chat. Only after the test, let it handle a live conversation, and read its drafts for the first few days.
What breaks: the browser plans stop when WhatsApp Web is closed or your computer is off. If the agent goes quiet, check that before you rewrite the instructions. Also, because the handoff rule is not published, a message that should reach a person may get an answer instead. You can step in yourself: open a conversation, and the Takeover button pauses the agent in that one chat while you handle it. There is also a setting, on by default for new agents, that pauses the agent in a chat as soon as you type a message there; it applies to new conversations. Test that case before you depend on the agent.
For how this fits next to a shared team inbox, the inbox management comparison lays out the two shapes side by side.
Where does your agent run, and what uptime do you actually get?
Your uptime depends on the plan you pick, not on the agent's instructions. Basic and Standard run in your browser and stop when WhatsApp Web is closed or your computer is off. Pro is cloud-hosted and keeps working 24/7. Standard does not include always-on sending.
| Plan | Where it runs | Keeps working when your computer is off? |
|---|---|---|
| Basic | Your browser | No |
| Standard | Your browser | No, and no always-on sending |
| Pro | Cloud-hosted | Yes, 24/7 |
Moving from Standard to cloud costs $25 per user per month. Every plan also carries a monthly AI-credit allotment, with allotments of 50, 100, and 200 credits across the three tiers. Build your uptime expectation from the table, not from the homepage headline.
If you run a scheduled job on a laptop, the job fires only while the laptop is awake. For the full set of what an agent can read and send, the WhatsApp MCP tools breakdown lists the actions.
What breaks, and where is the real risk?
Every route has a failure mode. Browser-based plans stop when the tab or computer stops. The official API limits you through tier caps and template review. A self-hosted server stops with your machine. The risk you control is who receives your messages.
Read-and-draft on chats that already exist is the low-risk end. Cold bulk sends to people who never saved your number are the high-risk end. No tool, Blueticks included, can promise zero risk there. The control that works everywhere is the same: message people who expect you, pace your sends, and keep a human approving anything outbound. The human-in-the-loop approvals guide shows how to set that up.
FAQ
Can I put ChatGPT or Claude directly on WhatsApp through the official API?
Not for a general-purpose assistant, according to TechCrunch's report of Meta's updated terms (effective January 15, 2026). Meta said customer-service bots are unaffected, so a business bot needs a defined job. Check Meta's current terms, since scope can vary by region. Running an assistant on your own account through an MCP server or the Blueticks agent is a different setup.
Does a Blueticks agent use the WhatsApp Business API?
On Basic and Standard, the connection runs on your computer through WhatsApp Web and the Blueticks extension, on your own WhatsApp account, not the Business API. On Pro it runs in the Blueticks cloud, 24/7. See the pricing page for plan details.
Can a WhatsApp AI agent get my number banned?
It can, if the way you use it breaks WhatsApp's rules. Risk tracks behavior. Reading chats and drafting replies you approve sits at the low end. Cold bulk sending sits at the high end. Treat every outbound message as something you would send yourself.
How much does a Blueticks agent cost?
Plans are priced per user per month across Basic, Standard, and Pro. Each plan includes a monthly allotment of 50, 100, or 200 AI credits. Cloud hosting, which keeps the agent running 24/7, is part of Pro, and moving from Standard to cloud costs $25 per user per month.
Is there a limit on how many messages my agent can send?
Yes. The Blueticks API page publishes per-number limits of 60 messages per minute on Basic, 120 on Standard, and 300 on Pro. The official API has its own limits, which start at 250 unique customers in a rolling 24 hours for a new portfolio.
Pick once, then build
Pick by job. A customer-facing bot at scale means the official API and its approval process. An agent that works for you on your own number means the Blueticks agent or a self-hosted MCP server, and the choice between them comes down to whether you need uptime while your computer sleeps.
Describe the agent in plain words, connect the number you already use, and send a test from a second phone before you let it near a live customer.
