linksmcp Request early access

The LinkedIn MCP guard that knows when to say not today.

For people running LinkedIn outreach with their own AI agent: a guard MCP server that checks every LinkedIn action against daily limits, pacing and pauses, on your own machine with Claude Code or any MCP client. Early access, by request.

Request early access See the rules

Request early access opens an email to [email protected]. Tell us which agent you use and how many profiles you run.

Runs on your machine. Your own LinkedIn session. Nothing sent anywhere by default.

Guard ledger
Your profile · today
  1. Connection requestMaya LinAllowedwait 42 s
  2. Connection requestTom BeckerAllowedwait 51 s
  3. 18 connection requestspaced across the morningAllowedpaced
  4. Connection requestAna SousaNot today20 of 20 used
  5. Message with a linkDana RuizRefusedno reply yet
  6. ReplyPriya NairAllowedreplies go
  7. Signed out by LinkedInoutreach pausedPaused48 h
Your agent asked for 30 connection requests and a follow up with a link. The guard let 20 through at a human pace, held the link until the lead replies, kept replies flowing and paused outreach when LinkedIn signed the profile out.

An agent with a LinkedIn tool has no brakes.

A LinkedIn MCP server does what your agent asks; nothing in it says not today. Ask for 200 invitations and it starts on the first one, at machine speed, at three in the morning, to someone another of your agents already wrote to.

linksmcp is the part that says allow, wait or no. It keeps your own AI agent inside conservative daily limits, pacing and pauses modelled on what keeps LinkedIn accounts healthy, and remembers what was sent so nothing goes twice.

LinkedIn tool on its own
  • InvitationSent
  • InvitationSent
  • InvitationSent
  • Link in noteSent
  • InvitationSent
  • Message, againSent
  • and on, until something stops it
Through the guard
  • Outside hoursNot now
  • InvitationAllowed
  • After 51 sAllowed
  • Link in noteRefused
  • After 3 minAllowed
  • Message, againRefused
  • and stops at today's limit

The guard sits between your agent and its LinkedIn tool.

The guard starts the LinkedIn tool itself, so your agent cannot go around it. A refused call never reaches LinkedIn.

Your agent
Claude Code, OpenClaw, Cursor or any MCP client
asks
The guard
Allow with a permit and a wait, or deny with a plain reason and a retry time
limitspacingpausesledger
permit
LinkedIn tool
The open source LinkedIn MCP server, or the Playwright MCP browser server
acts
Your browser
Your own LinkedIn session, on your own internet connection

Every action runs the same loop

  1. 1
    Preflight. The agent asks the guard first. The answer is allow, with a permit and a wait time, or deny, with a plain reason and, when time solves it, a retry time.
  2. 2
    Act through the guard. No permit, no LinkedIn action. The guard waits the pacing for you, then passes the call on.
  3. 3
    Record the outcome. What was sent, to whom, and any pushback from LinkedIn go into the ledger, so nothing goes twice.

What you install

The guard
A small MCP server between your agent and its LinkedIn tool. It refuses any LinkedIn action without a permit, waits the pacing and records what happened.
A LinkedIn tool
The open source LinkedIn MCP server by stickerdaniel (complete checks, pinned version), or the Playwright MCP browser server (best effort checks).
The skill
A Claude Code plugin with the loop the agent follows for every action.
Some of the plain reasons it gives
  • Outside today's working hours
  • Today's limit for this action is spent
  • Too many invitations pending
  • Less than 24 hours since your last automated message to this person
  • A link to someone who has not replied
  • LinkedIn pushed back recently
  • The first part of a session is deliberately slow
  • The person opted out, or another of your profiles is already talking to them
  • Profile name or photo is locked

The rules, with the real numbers.

Defaults of the default preset as of 2026-10-19. LinkedIn changes its behaviour without notice, so these are conservative starting points, not a promise.

Daily limits per profile default preset Every profile gets its own variation of limits and timing each day
Connection requests
20 per day

Today's cap lands between 70 and 100 percent of the limit, so somewhere from 14 to 20.

Messages
100 per day

Openers, replies and follow ups together. Replies to people who wrote to you are never cut.

Warm-up for a new profile

Connection requests reach the full 20 on day 16. 7 days idle restarts the warm-up.

Other actions, per day
Profile views250 1 block = 10 views
On one scale, 1 block = 1 action
  • Searches30
  • Accepting incoming requests20
  • Withdrawing old invitations20
  • Comments15
  • Reactions3
Fixed limitToday's cap lands in this band, 70 to 100 percent
Working hours
08:0014:0020:00

08:00 to 20:00 by default. Each day the window starts up to 45 min later and ends up to 45 min earlier.

Pacing between actions
  • Most waits 20 to 60 s
  • About one in three 2 to 5 min
  • A longer pause after 3 to 5 quick actions

Slow start after sign-in

LinkedIn calls per 10 minutes, for the first 6 hours after signing in.

Pauses after LinkedIn pushes back

48 hno outreach after a sign-out, security check, restriction or warning
+72 hcooldown on writes when LinkedIn ends the session itself

Replies to people who wrote to you still go.

Pending invitations

600pending at most, about a month of this profile's own sending at the default limit

Invitations pending 21 days or more are withdrawn, oldest first, in small batches.

And on every action

No links before a reply
A message, invitation note or comment with a link to someone who has not written back is refused, not edited. Comments and replies on your own posts may carry links.
24 hours between automated messages
To one person, while you were the last to speak.
Opt-outs are final
Someone who asks you to stop is never contacted again.
Profile edits
Name and photo are never changed by the agent. Other sections at most once per 24 hours.
Ordinary activity
A reminder every 2 to 3 days to spend a few minutes on the feed, a few reactions and notifications.
Your own profiles only
Never shared or bought ones. Your internet connection should match the profile's country.
Stricter preset for new profiles

New or recently warned profiles

10 connection requests50 messagesper day

A link waits until the person writes back.

Your agent can draft the follow up. The guard decides when it goes. A message, invitation note or comment with a link to someone who has not replied is refused, not edited, and the agent hears why.

  • okInside today's working hours
  • ok41 of 100 messages used today
  • ok26 h since your last message to Dana
  • heldA link before Dana has replied
MessagingYou
Dana RuizHead of Partnerships
Monday

You

Thanks for connecting, Dana. Curious how your team runs partner onboarding at the moment.

No reply from Dana yet
Draft from your agent

Following up with the case study I mentioned: yourdomain.com/case-study

Send

Held: Dana has not replied yet. It goes once she does.

Heldlink before reply

Local by default. Hosted when one profile has several agents.

Local mode keeps the ledger in a file on your machine. Hosted mode is optional, for two machines or several agents on one profile.

Local default Hosted optional
Where the ledger livesA file on your machineA server you connect with an API key and server address
What leaves your machineNothing. No account, no telemetry, no update check.Profile settings, permits, reported signals, lead identifiers (profile address, name, headline, company) and thread times
Message textStays on your machineStays on your machine. The guard sends a hash and the list of links; the server does not accept text at all.
Two machines or several agents on one profileEach machine keeps its own countCounted together
One lead, several profilesWithin one ledgerA person one of your profiles is talking to is not contacted by another
Server unreachable, erroring or over 8 sNot applicableThe guard refuses the LinkedIn action. It fails closed.

Will this stop my account getting restricted? No guarantee.

Here is what linksmcp will not do, in plain words, before you ask for access.

  1. 01

    It does not make LinkedIn automation safe.

    LinkedIn's User Agreement does not allow automated activity. You run the automation and you carry that risk. The limits lower the chance of a warning; they do not remove it.

  2. 02

    It only knows what your agent reports.

    The ledger trusts what your agent tells it. Actions you take by hand are invisible to it unless reported.

  3. 03

    It cannot check an agent that goes around it.

    An agent that calls the LinkedIn tool directly, around the guard, is not checked at all. That is why the guard starts the LinkedIn tool itself.

  4. 04

    Browser automation is best effort.

    With the Playwright MCP browser server the checks are best effort. The LinkedIn MCP server executor has complete checks.

  5. 05

    It never sees your LinkedIn login.

    No password, no cookie, and it never opens a LinkedIn page. Your agent does the work in your own browser.

Works with the agent you already run.

Agents
  • Claude CodeInstalls as a plugin with the skill
  • OpenClawAn MCP client, through a JSON config entry
  • CursorAn MCP client, through a JSON config entry
  • Any MCP clientThat can start a stdio server
LinkedIn tools the guard starts
  • LinkedIn MCP serverThe open source server by stickerdaniel. Pinned version, every tool checked by name and argument.complete checks
  • Playwright MCPBrowser automation.best effort checks
What setup looks like once you have access
MCP client config (JSON)preview
{
  "mcpServers": {
    "linksmcp": {
      "command": "node",
      "args": ["/path/to/guard.js"],
      "env": {
        "LINKSMCP_EXECUTOR": "stickerdaniel",
        "LINKSMCP_API_KEY": "",
        "LINKSMCP_HOSTED_URL": ""
      }
    }
  }
}
Leave the key and address empty for local mode.

The guard is the only LinkedIn server your agent sees; it starts the LinkedIn tool behind itself. In Claude Code it installs as a plugin with the skill. There is no public install yet: keys are handed out by hand.

Questions people ask first.

Anything else: [email protected]

What is a LinkedIn MCP server?

An MCP server that gives an AI agent such as Claude Code tools to act on LinkedIn: search, view profiles, send invitations and messages. It does what the agent asks. linksmcp is a guard MCP server that sits in front of it and checks every LinkedIn action against daily limits, pacing and pauses first.

Will this stop my LinkedIn account getting restricted?

No tool can promise that. LinkedIn's User Agreement does not allow automated activity, and you carry that risk. linksmcp keeps your agent inside conservative daily limits, pacing and pauses, which lowers the chance of a warning. It does not remove it.

Does linksmcp see my LinkedIn password?

No. It never sees a LinkedIn password or cookie and never opens a LinkedIn page. Your agent does the LinkedIn work in your own browser, with your own LinkedIn session and your own internet connection.

Which AI agents work with it?

Claude Code (as a plugin), OpenClaw, Cursor, and any MCP client that can start a stdio server, through a JSON config entry. For the LinkedIn side it works with the open source LinkedIn MCP server by stickerdaniel (complete checks) or the Playwright MCP browser server (best effort checks).

Do I need the hosted version?

Only if two machines or several agents work on one LinkedIn profile, or several of your profiles share leads. Local mode is the default: the ledger is a file on your machine and nothing is sent anywhere. In hosted mode message text still never leaves your machine, and if the server cannot be reached the guard refuses the LinkedIn action.

What are the daily limits?

Per profile, in the default preset as of 2026-10-19: connection requests 20, messages 100, profile views 250, searches 30, comments 15, accepting incoming requests 20, reactions 3, withdrawing old invitations 20. New profiles warm up from lower numbers, and a stricter preset for new or recently warned profiles allows 10 connection requests and 50 messages.

Is linksmcp affiliated with LinkedIn?

No. linksmcp is an independent tool and is not affiliated with, endorsed by or sponsored by LinkedIn Corporation.

How do I get access?

It is in early access and keys are handed out by hand. Email [email protected] with the agent you use and how many LinkedIn profiles you run.

Give your agent a guard before it gets busy on LinkedIn.

Request early access See the rules

Early access, by request. Runs on your machine. Your own LinkedIn session.

Guard ledgerYour profile · today · 11:06
Connection requestAna SousaNot today20 of 20 used
Reason
Today's limit for this action is spent
Retry
Tomorrow, inside working hours
Reached LinkedIn
No