The Instagram Claude sales engine is a five step outreach pipeline: find a business with a DM-worthy offer, scrape its likes and followers into a targeted list, run a Claude-driven iPhone farm, send personalized DMs from that business’s own account, then let Claude agents keep the conversation going until it converts. Kevin walks through the whole thing on KevBuildsApps.
niche + DM-worthy offer
|
v
scrape likes & followers -> targeted list
|
v
Claude-driven iPhone farm (many phones at once)
|
v
personalized DM from the target's own account
|
v
Claude agent keeps the thread warm
|
v
conversation converts (call booked, offer sold)What is an Instagram Claude sales engine?
It is a repeatable service that turns an Instagram account’s likes and followers into personalized DM conversations run by Claude, instead of by a human VA. The agent handles the volume; the offer and the account belong to the client.
The key structural point, and the one most people miss: the DMs go out from the client’s own Instagram account, not from a cold account you own. Followers already know the name in the inbox. That single detail changes reply rates more than any prompt tweak.
Three things have to line up for the engine to work:
- A target with real DM traffic. If the business does not live and die by its inbox, there is nothing to automate.
- A list built from people who already engaged. Likes and followers are warm by definition.
- An agent that remembers the thread. Step one is easy. Step five is where the money is.
Kevin’s framing in the walkthrough: this is the one winning workflow he would build right now if he were starting an AI agency. That is a claim about one operator’s opinion, not a benchmark. Treat it as a starting point to test, not a promise.
How the 5-step Instagram DM pipeline works
The pipeline is five concrete steps, and each one has a clear output you can hand to a client or a subcontractor.
- Pick a niche and find an Instagram business with a DM-worthy offer. The offer has to be something a DM can plausibly sell: a program, a community, a booking. If the sale needs a demo, this is the wrong client.
- Scrape their likes and followers to build a targeted list. You are harvesting engaged accounts, not emails from a bought list. The list stays tied to the source account.
- Run the iPhone farm so Claude handles many iPhones at the same time. Kevin’s video describes 100 iPhones in parallel. Each phone is a real device running a real session.
- Send personalized DMs straight from their Instagram account. Each message is written for the recipient, not blasted from a template.
- Keep the conversation going in the DMs with your agents until it converts. The agent answers follow-ups, handles objections and pushes toward the next step.
Step three is the part that makes the economics work. A human VA sends maybe a few hundred DMs a day and needs sleep. An iPhone farm does not care about hour six. That does not make it easy. It makes it an operations problem: device health, session hygiene, and a queue that never sends the same message twice.
If you want to see how Kevin builds these agent workflows end to end, the launch video is the place to start.
How to scrape Instagram likes and followers to build a targeted list
Scraping likes and followers gives you a list of people who already raised their hand for that niche, which is why it beats a scraped email list. But it is also the step with the most legal and platform risk, and you should say that to every client up front.
What the list actually is, in practice:
- Engagers. Accounts that liked a specific post are the warmest segment. They saw the offer and reacted.
- Followers. A wider net. Good for volume, weaker for intent. Segment these separately.
- Recent activity. An account that liked a post last week is worth more than one that liked something a year ago.
Kevin’s video links a separate resource for the scraping tooling rather than walking through it line by line, so the specifics live there. What matters for the sales engine is the output: a deduplicated list of Instagram handles, tagged by source and recency, that the DM step can consume.
Two rules I would set before writing a single message:
- Only list people tied to the client’s own audience. Likes and followers of that account. Not a competitor’s list, not a purchased file.
- Let people opt out cleanly. If someone says stop, the agent’s memory has to record it and the queue has to respect it. Build that in step two, not later.
Legality depends on jurisdiction, the source platform’s terms and what you do with the data. Instagram’s terms restrict automated collection. Get the client to sign off on the method, in writing, and consult a lawyer if you are selling this as a service. Nothing here is legal advice.
What is a Claude-driven iPhone farm and how does it send DMs?
A Claude-driven iPhone farm is a set of physical iPhones where Claude controls each device’s session, so one operator can run outreach across many phones at once. In Kevin’s build, Claude handles 100 iPhones simultaneously.
The architecture is more mundane than the name suggests, and that is the point:
| Layer | What it does |
|---|---|
| Devices | Real iPhones, each logged into the target account’s session, each with its own network path |
| Controller | Claude decides which phone gets which message next and what to say |
| Queue | Holds the scraped list, dedupes handles, tracks who was already messaged and who replied |
| Memory | Stores each conversation so a follow-up never restarts from zero |
| Guardrails | Rate limits, send windows, opt-out flags, human escalation on hot leads |
The reason to use real phones rather than a script is session trust. Instagram treats a device and its network as part of the account’s identity. A farm, run carefully, keeps that identity consistent. It is also expensive: hardware, SIMs, racking, power, replacements, and someone on call when a phone drops off Wi-Fi.
Do not sell this as risk-free. Meta’s rules on automation still apply, and an account can be restricted regardless of how the messages are sent. The honest pitch is throughput plus personalization, with a clear explanation of what happens if the account gets limited.
How to keep the DM funnel going with Claude agents until it converts
The follow-up is where an Instagram Claude sales engine earns its fee, because most deals close four or five messages after the first one. The agent has to know who it is talking to and what was already said.
Run the same message through the agent three times in your head and check that it lands differently each time. That is the test. A pipeline that sends “hey, saw you liked our post” to the same person twice is a pipeline that gets muted.
Creator OS is one place to run that loop without custom glue. The hosted MCP server exposes tools including list_conversations, send_message and create_automation, so an agent can read the thread and reply in context rather than guessing. The docs at Social Agents describe the harness pattern. To wire a Claude client up to the account at all, the setup walkthrough is in the MCP docs.
A working loop, in plain English, that you can hand to Claude:
1. Pull the last 24h of conversations with list_conversations.
2. For each thread where the last message is from the lead:
- Read the history.
- Draft a reply that references something specific they said.
- If they asked pricing, answer with the client's real number
and propose a call.
3. Skip anyone flagged opt_out. Never re-open a closed thread.
4. If the reply is 3+ messages deep and the lead is ready to buy,
hand to a human.That last line matters. Agents are good at volume and consistency. Humans are better at the awkward moment when someone asks a question the offer does not answer. A clean handoff beats a clever prompt.
For the comment-to-DM version of the same funnel, where a keyword in a comment triggers a DM, the API path is at the comments to DM use case. The KevBuildsApps channel has more walkthroughs of the agent side.
Can you auto reply Instagram DMs without getting flagged?
You can auto reply to Instagram DMs, but “without getting flagged” is not a promise anyone can make, and you should not repeat it to a client. What you can control is how human the pattern looks.
- Vary the message. Identical copy sent to hundreds of inboxes is the fastest way to a limit.
- Send inside normal hours. A burst at 3am is a tell.
- Respect the 24 hour window. Reply to people who just messaged you, not to cold handles you found last month.
- Warm the account first. Build the identity and the follower base before the farm starts sending.
- Have a stop switch. When a limit warning appears, pause the queue the same day.
Creator OS exposes DM sending through the API and the MCP server on the connected account, with read-only or read and write scopes. The write scope is what lets an agent reply on its own. The read-only scope is useful in a client demo, where you want to show the inbox without touching it. The same auth pattern applied to publishing is covered in scheduling Instagram posts with Claude.
Instagram MCP for Claude: how agents connect to the account
The Instagram MCP for Claude is a hosted endpoint a Claude client connects to, so an agent can read the inbox and reply on the connected account without custom glue. No install step, and the connection is scoped.
- Open Claude, go to Settings, then Connectors, then Add custom connector.
- Paste
https://mcp.creatoros.ca/mcp. - Sign in to Creator OS and pick the workspace. Agency plans connect one client workspace per connection.
- Choose read only or read and write.
Each connection gets its own key, listed under Settings, then Connected apps. Read-only connections only see read tools, and the API refuses writes from them. Destructive tools, including delete automation and delete webhook, are marked so the AI app asks before acting.
For Claude Code, the same server installs from the command line:
claude mcp add --transport http creatoros \
https://mcp.creatoros.ca/mcp \
--header "Authorization: Bearer cos_live_..."From there, tools like list_comments, reply_to_comment and create_automation give the agent the calls it needs for the funnel. The comment-to-DM automation is where the sales engine overlaps with publishing: a keyword in a comment starts a DM, the agent continues the thread, and the deal closes in the inbox. The full tool list is at the MCP docs.
One limit worth knowing before you promise anything: Creator OS handles comment-to-DM funnels on Instagram and Facebook. TikTok is not part of that funnel. Plan the client offer around the platforms the funnel actually covers.
Who should you sell this Instagram sales engine to first?
Sell it first to online coaches who run a one-on-one program and a community offer. They already live in the DMs, they have a clear offer, and they can see the value of a warmer inbox in one week.
Why coaches are the cleanest first client:
- The offer fits in a message. “Book a call about the program” needs no demo.
- The audience is already on Instagram. Their followers are the list.
- The conversion event is visible. A booked call is easy to count, which makes your first invoice easy to justify.
After the first two coaches, the pattern extends to anyone whose business is a DM conversation: course creators, agencies with a booked-call model, and service businesses with a clear next step. What it does not fit is a business whose sale needs a demo, a proposal or a long procurement. Those belong in a pipeline with a human on the other side from message one.
Price it on outcomes you can actually count: conversations started, replies received, calls booked. Not on messages sent. Sent volume is your cost, not the client’s result.
FAQ
What is an Instagram Claude sales engine?
It is a five step outreach pipeline where Claude finds a target account, works from its likes and followers, sends personalized DMs from that account, and keeps the conversation going until it converts. The work runs inside the client’s own Instagram presence, not from a cold account you own. Kevin from KevBuildsApps walks through the full build in his video.
Can Claude send Instagram DMs automatically?
Yes, when Claude is connected to an account through the hosted Creator OS MCP server with read and write scope. The agent calls tools like list_conversations and send_message to read a thread and reply. Each connection gets its own key, and read-only connections cannot write.
How does the claude instagram mcp work for DM outreach?
You add https://mcp.creatoros.ca/mcp as a custom connector in Claude, sign in, and choose the workspace and the scope. The agent then gets tools for the inbox, comments and automations, so it can read a thread and respond without custom code. The setup steps are in the MCP docs.
Is scraping Instagram likes and followers legal?
It depends on your jurisdiction, Instagram’s terms and how the data is used, and Instagram’s terms restrict automated collection. Get written sign-off from the client on the method and take legal advice before selling it as a service. Nothing in this article is legal advice.
How do I auto reply to Instagram DMs with AI?
Connect the account to an agent through the MCP server or the REST API with a write-scoped key, then give the agent rules for tone, opt-outs and escalation. Keep messages varied, send inside normal hours and pause the queue when a limit warning appears. Creator OS provides the connection and the tools; your agent’s prompt decides what it says.
Get started
Creator OS plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included. Connect the Instagram account, add the connector, and run the five step loop on one client before you build the farm.
Sign up at creatoros.ca. Then read the MCP docs for the connection details and the Social Agents docs for the harness pattern. Watch the walkthroughs on the KevBuildsApps YouTube channel and the launch video to see the agent side in action.
Get started for $0
Free trial for new subscribers. Cancel anytime.
