A Claude iPhone farm is a setup where Claude drives many iPhones in parallel to run marketing tasks: text blasts, social posting, Instagram cold DMs and LinkedIn outreach. Kevin built an open source version in KevBuildsApps that controls up to 100 iPhones at once, one workflow per device. This post walks through how it works and how to install it from the source video.
Claude (your prompt)
|
v
iPhone farm controller (open source, KevBuildsApps)
|
+-- iPhone 01 text blasts / TikTok / Instagram / LinkedIn
+-- iPhone 02 text blasts / TikTok / Instagram / LinkedIn
+-- iPhone 03 ...
+-- iPhone 100 (up to 100 devices in parallel)
|
v
Per device: warm up, post, scrape, DM, connect
Per social network: up to 10 accounts on one phoneThe project is Kevin’s own open source work, published under KevBuildsApps. The walkthrough lives here: the iPhone farm video. Watch it once before you install anything, because the install order matters and the video shows the exact sequence.
What Is a Claude iPhone Farm?
A Claude iPhone farm is a pool of real iPhones that Claude can operate at the same time, instead of you holding one phone and tapping. The phone stays the client, so the apps see a normal device. Claude is the operator, so the workflow is written in plain language.
Kevin’s build targets 100 devices in parallel. The description gives the working numbers: each iPhone can hold up to 10 socials per network, so a 10 phone farm can push a thousand uploads a day. Those are capacity figures from the project author, not a promise about what your accounts will do.
What you actually get from one prompt:
- A text messaging tool for blasts to leads and customers.
- TikTok and Instagram warm up, then posting at scale.
- Instagram profile scraping and personalized cold DMs.
- 25 LinkedIn connection requests per day, autonomously, on each phone.
The reader this fits is a solo operator or a small agency that needs outbound and content volume without hiring a team. It is not a replacement for a scheduler. If you want one dashboard for publishing, replies and analytics across 11 platforms plus Skool, that is a different job, and Creator OS does that part from the web app, the iOS app, the API, the CLI and a hosted MCP server. Use the farm for device level outreach and Creator OS for the account level work.
How Does the Open Source iPhone Farm Setup Work?
The farm is a controller plus a fleet: Claude sends instructions, the controller routes them to a free iPhone, and the phone performs the action in the real app. Each phone is a separate lane, so a failure on device 37 does not stop device 38.
The division of labour in the video is worth copying into your own plan:
| Layer | Job | What you own |
|---|---|---|
| Claude | Reads your prompt, decides who gets what message | The prompt and the lead list |
| Controller | Assigns work to iPhones, tracks what finished | The install and the device list |
| iPhone | Runs the messaging app, TikTok, Instagram, LinkedIn natively | The device, the SIM, the accounts, the network |
| Your data | Profiles, handles, conversation history | Where it is stored and who can read it |
Two things make or break a farm. First, one account per phone per lane, because the whole point is that no single device looks like a spam cannon. Second, a queue you can read. When 100 phones are working you will not watch them. You will read a summary and spot the lanes that stalled.
A worked example of a prompt that fits this structure:
Use phones 1 to 20.
For each phone, open Instagram and send a DM to the
handles in leads-day1.csv, one handle per phone per hour.
Personalize the first line from the profile bio.
Stop after 8 sends per phone.
Report: sends per phone, replies, and any phone that
hit a rate limit.Note what the prompt does not do: it does not ask for unlimited sends. You set the ceiling. The project gives you the hands, not the judgement.
How to Install the iPhone Farm Project and Connect Claude
Install comes first, then device pairing, then a single test phone before you scale. The video description lists the install as step one: install the open source project that lets Claude control up to 100 iPhones in parallel. Follow the repo instructions for the exact commands rather than guessing at flags, and watch the KevBuildsApps walkthroughs while you do it.
The order that works:
- Install the project and confirm it starts.
- Connect one iPhone. Run one harmless action on it, like opening an app.
- Add three phones. Run the same action on all three.
- Only then add the rest, in blocks of 10, checking the report after each block.
If you are new to giving Claude tools and permissions, read the open source iPhone farm and iOS automation post first. It covers the same project and explains the moving parts without the outreach angle.
On the Creator OS side, the work that does not need a phone runs through the hosted MCP server at https://mcp.creatoros.ca/mcp. Claude Code users can add it with:
claude mcp add --transport http creatoros https://mcp.creatoros.ca/mcp --header "Authorization: Bearer cos_live_..."Every connection gets its own key and can be read only or read and write, listed under Settings > Connected apps. That keeps a marketing agent inside one workspace. The setup details are in the MCP docs.
How to Send Text Blasts to Leads and Customers
The farm ships a built in text messaging tool, so a blast is a prompt plus a list, not a manual tap session on 40 phones. Kevin’s description names it directly as step two: use the built in tool to send text blasts to leads and customers.
The difference between a blast that gets replies and one that gets blocked is in the list, not the tool:
- Split customers from cold leads. Customers get updates and offers. Cold leads get one opener and nothing else until they answer.
- Send in waves of a few phones, not all 100 at once. Same total, softer curve.
- Write the first line so it could only be sent to that person. If it fits anyone, rewrite it.
- Keep a stop list and honour it the same day.
Example prompt for a customer wave:
Use phones 1 to 5.
Send the message in customers-restock.txt to everyone
in customers.csv who has not been messaged this week.
One message per number. Do not follow up.
Log number, phone, timestamp, and any send error.Keep this separate from Creator OS. The farm’s text messaging runs on your own phones and SIMs. Creator OS handles WhatsApp Business messaging, which is a different channel with its own rules: free text only within 24 hours of the customer’s last message, templates outside that window. See the WhatsApp API docs if you run both.
How to Warm Up and Post to TikTok and Instagram at Scale
Warm up is not optional at farm scale: a new account on a fresh phone that posts the same clip as 40 other accounts is the fastest way to lose the whole batch. Treat the first days as training, not publishing.
A warm up sequence that matches the project’s flow, step three in Kevin’s description:
- Days 1 to 3: sign in on the account, follow a few people in the niche, watch videos for real minutes. No posts.
- Days 4 to 7: one post per account, different clip per account, different caption.
- Week 2 onward: raise to a normal cadence, still one account per lane per phone.
The capacity math from the project: up to 10 socials per network per iPhone, so a 10 phone farm can push a thousand uploads a day. That is a ceiling for the hardware, not a target for week one. Volume only helps if the accounts survive.
For the account level side, Creator OS is where the posts live once they are worth keeping. A scheduled post through the CLI is one line:
creatoros posts:create --text "New drop is live" \
--platforms instagram,tiktok \
--media https://example.com/clip.mp4 \
--scheduledAt 2026-03-02T18:00:00Z --timezone America/TorontoCreator OS applies each platform’s own rules for Reels and TikTok consent flags, and it reports networks that are not connected back in missing_platforms while the rest still go out. Post options are in the post options docs.
How to Scrape Profiles and Send Personalized Instagram Cold DMs
Scraping plus a personalized first line is the whole trick; scraping plus a template is spam with extra steps. The project’s step four is exactly this pair, in order.
What “personalized” has to mean in practice: one true detail from the profile, used in a sentence that would read wrong on anyone else. A bio mention, a recent post topic, a city. If the scrape does not return something specific, skip the lead.
Use phones 1 to 10.
For each profile in ig-leads.csv, open the profile,
capture bio, follower count and latest post topic.
Draft a DM with one specific line from that data,
then a one sentence ask.
Hold all drafts. Do not send until I approve.Approval before sending is the part people skip and regret. If you already use Claude Code, the approval pattern is the same one described in AI agent approvals on your iPhone lock screen: the agent pauses, you decide, the work continues.
Keep your DM volume inside what one account can do in a day. The farm lets you spread work across devices; it does not make a single account look human when it messages 300 strangers before lunch.
Can You Run LinkedIn Outreach on Autopilot?
Yes, within the limit the project sets: 25 LinkedIn connection requests per day per phone, autonomously. That is the number in Kevin’s description, and it is the right kind of number, because it is per device and per day.
With 10 phones that is 250 requests a day across the farm, with each phone’s account staying inside a normal daily rhythm. The prompt that keeps it honest:
Use phones 1 to 10.
For each phone, send up to 25 LinkedIn connection
requests today using the note in li-note.txt.
Skip anyone already connected. Skip anyone we
messaged in the last 90 days.
At 6pm, report requests sent per phone and
acceptance count.Autopilot here means “runs without you tapping”, not “runs without you reading”. Check the report daily. If a phone’s acceptance rate drops, that note is not landing and you should change the note, not the volume.
Claude Marketing Skills That Power the Farm
The farm gives Claude hands on phones. Skills give it the marketing playbook, so the same prompt produces a captioned post instead of a raw file upload.
Creator OS exposes marketing skills two ways. Locally, npx @creatoros/cli@latest init installs 14 Claude Code skills, including post-shortform, respond-to-dms, automations, analytics, ads and the four campaign builders. Over MCP, the same skills arrive as prompts, so they show up in the AI app’s slash menu. Read the full list in the AI OS skills post.
Tool names worth knowing when you wire the two stacks together: create_post, check_caption_length, send_message, get_best_time_to_post, get_post_analytics. About 160 tools exist in total, and destructive ones such as delete post or delete ad are marked so the AI app asks first.
If you would rather run an agent that handles the socials without a phone farm, Social Agents is another of Kevin’s open source projects: start with npm start creatoros social-agents, bring your own model, and it interviews you about your brand before it posts. See the lead generation post for that workflow, and watch the walkthroughs on the KevBuildsApps YouTube channel. The launch video for Creator OS is here.
FAQ
What is an iPhone device farm and how does Claude control it?
An iPhone device farm is a group of physical iPhones that software can drive in parallel. In this project, Claude sends the instructions, the controller routes each task to a phone, and the phone performs it in the real app. That is why the scale numbers are per device: one phone, one lane, one account.
Is there an iPhone farm GitHub repo I can use for free?
Yes. It is Kevin’s open source project, described in the walkthrough at the iPhone farm video on KevBuildsApps. Follow the installation steps shown there and in the project’s own instructions rather than relying on a summary. It is free to install; the phones and accounts are your cost.
How many social accounts can one iPhone farm run?
The project description says each iPhone can hold up to 10 socials per network, and the setup targets up to 100 iPhones in parallel. So the ceiling is large, but it is a ceiling for hardware, not advice to fill it on day one. Start with one phone, then add in blocks of 10.
Can Claude really send marketing texts and cold DMs without getting banned?
No one can promise that, and you should be suspicious of anyone who does. What the project does is let you spread work across many devices and set per phone, per day limits, which is how a real operator keeps volume normal. Deliverability and account safety depend on your lists, your copy, your opt outs and each platform’s rules.
What Claude skills for marketing do I need to run this stack?
For the phone work, the prompts in the video are enough to start. For the account level work, run npx @creatoros/cli@latest init to install 14 Claude Code skills including post-shortform, respond-to-dms, analytics and ads. They are included in every Creator OS plan, and the same skills arrive as MCP prompts if you connect the hosted server instead.
Get started
Build the farm small: install the project, pair one iPhone, run one prompt, read the report. Then add phones in blocks of 10 and keep a written ceiling per device per day. That habit matters more than the hardware count.
For the half of marketing that should not run on a phone, sign up at https://www.creatoros.ca/sign-up and connect the accounts you want Claude to manage directly. Creator OS plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included. Agency plans start at $49.99/month for 5 client workspaces. New subscribers start for $0 with a free trial and can cancel anytime.
Get started for $0
Free trial for new subscribers. Cancel anytime.