linkedin-mcp

Governed LinkedIn hands for AI agents. This is the public face of Apex by LeadShark — an MCP loadout that lets Claude, ChatGPT, or Cursor operate a real LinkedIn account in plain English: find buyers, rank intent, comment, connect, and message. Every action is paced, capped, approval-queued where needed, and fully logged.

This repo contains the discovery endpoint — a minimal, dependency-free MCP server, served live at:

• https://www.apex.new/mcp (alias: https://www.apex.new/linkedin-mcp)

Mount it in any MCP host and it completes the handshake instantly — no account, no auth — then briefs your agent on what the full loadout is and how to get it.

The mission

Most LinkedIn tooling automates what you already decided: schedule this post, DM whoever comments. Apex is the layer above that — the intelligence that decides what to automate and where to show up. The activation path is intelligence-first, not automation-first. In five words:

Discover demand. Engage with context.

An agent running Apex works a loop: See → Decide → Act. Find the posts and people where attention already is, figure out who's worth engaging and with what angle, then engage — inside the account's safety budget.

The loadout — 35 tools, See → Decide → Act

The full authenticated server at https://apex.leadshark.io/mcp equips your agent with:

See / discover — where is the demand?

search · discover_lead_magnets · list_signals · feed · list_recent_posts · list_person_posts · list_post_engagers · get_person_activity

Find people and companies, pull a continuously refreshed radar of prospect-rich lead-magnet posts, list your heat-scored inbound signals, turn one viral post into a warm lead list, and see the rooms a prospect is already in (get_person_activity — who they comment on, what they react to).

Decide / context — who is worth your time?

enrich · enrich_company · get_lead_activity

Structured profiles for a person or account: role, company, activity, connections. The agent ranks ICP fit and intent and picks the angle — comment, connect, or ignore. Built on enrich, the most-used signal in the loadout.

Act — engagement

comment_on_post · react_to_post · send_connection_request · engage_with_comment · manage_invitations

Show up where the prospect's attention already is — warm them up in public before anything lands in their inbox. Triage inbound connection requests too.

Act — messaging

list_recent_messages · get_messages_with_person · send_message

Read the inbox, pull a full thread with one person, send the DM — deduped, paced, and logged.

Company Pages (Pro+ baseline)

list_companies · comment_as_company · react_as_company

Engage in your Company Page's voice on pages you administer.

Automations & scheduled posts (Pro+ baseline)

create_automation · list_automations · get_automation · edit_automation · suggest_automation_settings · schedule_post_with_automation · list_scheduled_posts · get_scheduled_post · edit_scheduled_post · cancel_scheduled_post

The execution layer: comment→DM lead-magnet automations and pre-automated post scheduling. Apex decides what to run; these run it.

Safety, limits & history

manage_activity_limits · set_daily_dm_limit · list_actions

Check budgets before acting, adjust caps, and read the audit trail — what the agent already did, what's staged for approval, why something was refused, and the current safety mode.

Full per-tool reference (parameters, access, tiers): apex.new/tools.md

The safety model (this is the point)

Every write the agent makes shares the same daily/weekly budgets as the account's automations (DMs, replies, connections), honors the do-not-engage list, and is paced and rate-limit-aware. Actions can be refused or queued for the user's approval — and the agent is told why. Limits are calibrated from LeadShark's production LinkedIn volume, not guesswork. Agents are expected to check list_actions before acting and to work within limits rather than around them.

What this repo's server does

This discovery stub speaks MCP over Streamable HTTP (JSON-RPC 2.0 over POST, JSON responses, stateless, no SSE) and exposes three no-auth informational tools:

It has no LinkedIn powers of its own — it's the handshake and the briefing, not the hands.

Try it

Add https://www.apex.new/mcp as a custom connector / MCP server in your AI app, or run this repo locally:

node server.mjs          # Node 18+, zero dependencies, listens on PORT (default 3000)

Handshake by hand:

curl -s -X POST http://localhost:3000/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'

Get the full loadout

1. Create a LeadShark account — free 7-day trial, no card.
2. Connect the LinkedIn account the agent will operate.
3. Unlock the one-time 24-hour Apex operator window (Settings → Apex Settings).
4. Add https://apex.leadshark.io/mcp to your AI app and authorize with LeadShark.
5. Ask: "Check if you can see LeadShark Apex and list the tools available."

Apex is $119/mo after the trial. LeadShark Pro ($39/mo) automates your lead magnet; Pro+ ($59/mo) scales automations and gated funnels; Apex is the operator loadout on top.

• Website: apex.new · Pricing: apex.new/pricing
• MCP setup guide: apex.leadshark.io/docs/mcp
• Agent operating guide: apex.new/agents.md
• Machine-readable index: apex.new/llms.txt

License

MIT