Watches the people you name
You add leads by LinkedIn profile URL. The agent monitors those profiles and only those. It does not go looking for anybody else.
LinkedIn AI agent · one job
A LinkedIn AI agent with one job: show up every time your leads post.
Plenty of agents promise to run your whole LinkedIn. This one does a single thing and keeps doing it. When somebody on your list publishes, Icebreakly drafts a comment in your voice and posts it on your approval or on Autopilot. It sends no invites and no messages. Weeks later, the people you care about know your name.
- Checks chosen leads every day
- Comments in your Voice
- Approval or Autopilot
- 63 MCP tools for Claude
- No invites, no DMs
What is a LinkedIn AI agent?
A LinkedIn AI agent is software that watches for something on LinkedIn, uses a language model to decide what to do, and then acts on your account. The acting is what makes it an agent. A tool that gives you text to paste is an assistant.
Scope is the big choice. A broad agent searches, invites, messages and replies, and every extra power is one more thing that can go wrong under your name. Icebreakly is a narrow agent by design. It comments on posts from people you picked. That is the full extent of what it does on LinkedIn.
A day on shift
One invented lead, followed from a single post to the day she is ready to hear from you. The days are illustrative. The behavior is how the product works.
- 1
Monday. A lead posts
Marta, a VP of Operations on your list, publishes a post about a warehouse system rollout that went sideways. You are in a meeting and have no idea.
- 2
By Tuesday. The agent notices
Icebreakly checks each person you chose on a daily rotation and picks up her post on its next check. Nobody on your team had to open her profile.
- 3
Next. It drafts
A comment is written against what Marta said, in your Voice: your tone, your usual length, inside your guardrails. The draft is labeled Touch #7, because it would be your seventh touch with her.
- 4
Then one of two things happens
On Autopilot the comment is posted, and the post is liked too if you have that switched on. In Approval mode the draft waits under “Needs your review” until you approve it, change it or skip it.
- 5
Later that week. It holds back
Marta posts twice in one day. The agent comments once. About one comment per lead per day is the pace, with no bursts, because nobody replies to everything a person writes.
- 6
Some weeks on. It tells you she is ready
Marta passes 12 touches and appears on the warm-and-ready list. Sending the invite and writing the message are outside the agent’s job. Those are yours.
What the agent does, and where it stops
Six things it does. After them, a plain list of what it leaves alone: finding leads, sending invites, sending DMs or InMail, running sequences, managing an inbox, syncing to a CRM. If you need those, pair it with a tool built for them or choose a broader platform.
Writes in your Voice
Each draft answers the post in front of it and follows the Voice you trained: tone, length, style notes, language, emoji preference, guardrails and your own sample comments.
Posts with or without sign-off
Approval mode puts every draft in a queue for you. Autopilot, on Standard and Pro, posts inside your guardrails. You can change mode at any time.
Keeps score
Every lead has a touch count and a stage: Stranger, Familiar at about 5 touches, Trusted at about 12, Warm at 20 or more. The dashboard shows posts found, comments posted and posts liked, for today and all time.
Paces itself like a person
Roughly one comment per lead per day, from a dedicated cloud browser session for each LinkedIn account. You can attach your own proxy to an account if you prefer.
Works a team’s accounts
Pro includes three LinkedIn accounts and a shared workspace, so colleagues or agency clients run comment campaigns side by side. Slack notifications keep you posted without opening the app.
Run the agent from Claude
Icebreakly publishes an MCP server at mcp.icebreakly.com with 63 tools. Connect Claude, or any other MCP client, and manage the agent in plain language. Your client asks you to confirm before a comment or a like is posted. The full tool list and the Claude setup walkthrough have the details.
| You say | Tool the client calls | What happens |
|---|---|---|
| “Set up a campaign that watches my target accounts.” | create_comment_campaign | An always-on comment campaign is created with a Voice and a schedule |
| “Start watching these eight profiles.” | add_monitor_targets | The profiles join the monitored list |
| “What is waiting for my approval?” | get_pending_comments | Each pending draft comes back with its touch number |
| “Approve the first two.” | approve_pending_comment | Those comments are queued to post |
| “Let me hear this Voice on a few real posts.” | test_comment_voice | Sample drafts are written and nothing is posted |
| “Switch that campaign to Autopilot.” | update_comment_settings | The mode changes. Liking posts is set here too |
| “Who is warm?” | get_warming_leads | Monitored leads are listed by stage and ranked by touches |
Put the agent to work in four steps
Setup is short because the job is small. The slow part is the weeks of showing up, and the agent does that.
- 1
Connect a LinkedIn account
Sign up and connect the account the comments should come from. It runs in its own dedicated cloud browser session. If you would rather route it through your own proxy, add one for that account.
- 2
Name the leads
Create a campaign and paste in the profile URLs of the people you want to know you. The trial takes 10. Choose people who post, since a silent profile gives the agent nothing to do.
- 3
Give it a Voice and a mode
Describe your tone, set a length, add sample comments and write the guardrails. Start in Approval mode so you see every draft. The trial is Approval only, and paid plans add Autopilot.
- 4
Check in, do not babysit
Clear the review queue once or twice a day, and look at the warm-and-ready list once a week. If you live in Claude, connect the MCP server and do both from a conversation.
Give the agent its first ten leadsFree for 7 days in Approval mode.
LinkedIn AI agents: what to know first
If you are comparing options, see the LinkedIn AI SDR that warms leads, running comment campaigns from Claude, and the LinkedIn automation safety guide.
What is a LinkedIn AI agent?
A LinkedIn AI agent is software that observes something on LinkedIn, decides what to do about it using a language model, and then takes an action on your account without being told each step. That separates it from a writing assistant, which only hands you text. Agents differ widely in scope. Some try to run a whole outreach program: searching, inviting, messaging and replying. Others are built to do one task repeatedly and well.
What does the Icebreakly agent actually do?
One job. It checks the LinkedIn profiles you add every day, picks up each lead’s latest post, drafts a comment in your voice that responds to that post, and either waits for your approval or posts it on Autopilot. It can like the post as well. It counts every touch and tracks each lead from Stranger to Warm, so you can see who is ready to hear from you.
Will the agent send connection requests or messages for me?
No. It does not send connection requests, DMs, InMail, follow-up sequences or email. It does not find leads, scrape lists or look up contact details, and it has no inbox for managing replies. If you want an agent that runs the full outreach motion, a broader outreach platform is the right choice. Icebreakly is for the stage before outreach, when the aim is to become a familiar name.
Can the agent run without me?
On Standard and Pro, yes. Autopilot posts comments on its own inside the guardrails written into your Voice, at about one comment per lead per day. The free trial is Approval mode only, which means you see and approve every draft. Most people are better off starting in Approval until the drafts consistently sound like them, and you can switch between the two modes whenever you want.
Can I control it from Claude or another AI assistant?
Yes. Icebreakly runs an MCP server at https://mcp.icebreakly.com with 63 tools. An MCP client such as Claude can create comment campaigns, add or remove monitored leads, list and approve pending comments, read warming stages, tune a Voice and pull dashboard numbers. Access is scoped to the workspace behind your key, and it is included with your plan.
How does the agent avoid behaving like a bot?
By doing less. It comments about once per lead per day, never in bursts, and it acts through a dedicated cloud browser session for each LinkedIn account and not through raw API calls. You can add your own proxy per account. No tool that acts on a LinkedIn account is free of risk, and Approval mode is there if you want a person to check every comment before it is public.
Does it work across several LinkedIn accounts?
Yes. Pro includes three LinkedIn accounts, unlimited campaigns and a shared team workspace, and each further account is $40 a month. A sales team can warm the same target accounts from different reps, and an agency can run separate comment campaigns for separate clients from one place.
How much does the agent cost?
The first week is free: a 7-day trial limited to 10 leads and Approval mode. To let the agent run on Autopilot you need a paid plan. Standard, at $49 per month, covers 100 leads on a single LinkedIn account. Pro, at $99 per month, covers 300 leads on up to three accounts and adds the team workspace.
Hire the agent that only does one thing.
Hand it ten names and a Voice. It will be there the next time each of them posts, and the time after that. See plans and limits.