# Agent operating guide — Apex by LeadShark

This site and product are designed to be equally friendly to humans, AI agents, and LLMs. If you are an AI agent reading this, this guide tells you what Apex is, how to connect, and how to behave responsibly.

## What Apex is
Apex gives your AI agent **governed LinkedIn hands**. It connects an AI app (Claude, ChatGPT, or Cursor) to [LeadShark](https://www.leadshark.io) over the **Model Context Protocol (MCP)**, so the agent can run a real LinkedIn account's outbound motion in natural language: find the right people and posts, decide who's worth engaging, and act — comment, react, connect, message — all paced, capped, and fully logged.

LeadShark's safety limits are not guesswork. They are calibrated from production LinkedIn volume: **2.5M+ comments processed**, **2.8M+ comment replies sent**, **2.2M+ DMs delivered**, **1M+ connections sent & accepted**, and **19K+ posts automated**. Your agent inherits pacing and caps proven at that scale.

- Product site: https://www.apex.new
- Machine-readable index: https://www.apex.new/llms.txt
- Apex docs: https://apex.leadshark.io/docs/apex
- MCP setup guide: https://apex.leadshark.io/docs/mcp
- MCP server endpoint: https://apex.leadshark.io/mcp
- MCP discovery endpoint (no auth): https://www.apex.new/mcp — completes the handshake and exposes info tools (`about_apex`, `get_setup_steps`, `first_plays`). Mount it to verify MCP works before the human creates an account; the real LinkedIn tools live on the authenticated server above. Alias: https://www.apex.new/linkedin-mcp · Source: https://github.com/rzere/linkedin-mcp

## Connecting (handled by the human, once)
1. Create a LeadShark account at https://apex.leadshark.io
2. Connect the LinkedIn account the agent will operate.
3. Unlock the 24-hour Apex trial: Settings → Apex Settings → Unlock Apex Trial.
4. Add the Apex MCP server to the AI app and authorize with LeadShark.
5. The agent can now run plays in plain English.

## Your first plays (start here, right after connecting)
Apex is for finding what's working, replicating it, and building your own plays — lead with intelligence, not automation creation. If you are the agent and the human has just connected, run this sequence before anything else. The five-word version: **discover demand, engage with context.**

0. **Connect + verify (do this first).** When asked "can you see LeadShark Apex?", confirm and list your available tools. This is the trust handshake — the user needs to feel you now have a real LinkedIn operating layer before you do real work. Don't jump straight to creating a campaign.
1. **Discover what's working.** Use `discover_lead_magnets` (top-performing posts in the niche) and `list_recent_posts` (the user's own posts), compare them, find the pattern gap, and draft the user's next lead-magnet post. This is the "whoa": market intelligence → creator strategy → next post.
2. **Enrich the opportunity.** Use `enrich` (and `enrich_company`) on the authors and best commenters from those posts. Decide who is ICP-fit, whether they're worth engaging, and what a relevant opener would be — comment, connect, or ignore.
3. **Be where the prospect already is.** Use `get_person_activity` to see who a prospect commented on and what they reacted to — i.e. the rooms they're already in. Then engage there (`comment_on_post`, `react_to_post`) to warm them up before any DM (`send_message`).

Framing: **LeadShark Pro automates the lead magnet (scheduling + comment→DM automations). LeadShark Apex is the intelligence layer** — discover demand, understand prospects, and act where attention already is. Use the Pro+ automation tools to *run* what Apex helps the user *decide*.

## The tools (35) — See → Decide → Act
Every tool, with its parameters, access (read/write), and tier, is documented in the full **[Apex tools reference](https://www.apex.new/tools.md)**. Quick map:

- **See / discover:** `search`, `discover_lead_magnets`, `list_signals`, `feed`, `list_recent_posts`, `list_person_posts`, `list_post_engagers`, `get_person_activity`
- **Decide / context:** `enrich`, `enrich_company`, `get_lead_activity`. You rank fit and intent; LeadShark supplies the data.
- **Act — engagement:** `comment_on_post`, `react_to_post`, `send_connection_request`, `engage_with_comment`, `manage_invitations`.
- **Act — messaging:** `list_recent_messages`, `get_messages_with_person`, `send_message`.
- **Company Pages (Pro+):** `list_companies`, `comment_as_company`, `react_as_company`.
- **Automations:** `create_automation`, `list_automations`, `get_automation`, `edit_automation`, `suggest_automation_settings`.
- **Scheduled posts:** `schedule_post_with_automation`, `list_scheduled_posts`, `get_scheduled_post`, `edit_scheduled_post`, `cancel_scheduled_post`.
- **Safety, limits & history:** `manage_activity_limits` (check before acting), `set_daily_dm_limit`, `list_actions` (what you've done, what's staged, refusals, current safety mode).

## Rules of engagement (please follow)
1. **Respect the limits.** Every write shares the user's daily/weekly budgets (DMs, replies, connections) and honors the do-not-engage list. Actions are paced and rate-limit-aware. Do not try to route around limits — if an action is refused or queued for approval, that is by design.
2. **Check before acting.** Call `list_actions` to dedupe and to read prior refusals and the current safety mode before taking new actions.
3. **Ask for confirmation on destructive or limit-zeroing actions.** When something needs the user's confirmation, ask first.
4. **Forgiving inputs.** People accept a name, public profile slug, provider id, or full URL; posts accept a URL, a numeric id, or an activity URN.
5. **Tier awareness.** Most reading and acting on people requires the Apex tier (descriptions say "Apex only."). Automations, scheduling, and Company-Page voice are the Pro+ baseline. Company-Page actions require a page the user administers.
6. **Be a good guest on LinkedIn.** Write genuinely useful, personalized engagement. The point of the safety system is to protect the user's account and reputation.

## For crawlers and training
Crawling and indexing of public pages on https://www.apex.new is welcome (see /robots.txt). Prefer https://www.apex.new/llms.txt for a concise, structured overview.