MCP + LinkedIn: how AI agents automate outreach

Model Context Protocol (MCP) is an open standard that lets AI agents use external tools. Think of it as USB-C for AI - a universal connector that works across models and platforms. Claude, ChatGPT, and dozens of other AI tools already support it.

Crispy's live MCP catalog has 164 live MCP tools. A connected client runs the tools its permission scope allows. Each tool maps to a specific LinkedIn action: searching for people, viewing profiles, sending connection requests, writing messages, creating posts, and more.

Here's a real workflow: you tell Claude 'Find 10 marketing directors at Series B startups in Berlin and send each a personalized connection request.' Claude calls search_people to find matches, get_profile to read each person's background, and send_invitation with a custom note referencing their specific work. The entire sequence runs in under a minute.

What makes this different from traditional automation tools? Three things. First, the AI agent decides the strategy. You describe the goal, and it figures out the steps. Second, personalization is genuine. The agent reads each profile and writes unique messages. Third, Crispy enforces per-category daily caps. The defaults include 15 invitations and 150 messages. LinkedIn's User Agreement section 8.2 prohibits unauthorized automation. The caps do not authorize activity and do not guarantee against restriction.

For sales teams, this means one person with an AI agent can automate research and personalization while using the data-access and deletion options described in the Privacy Policy.

We're seeing teams run entire outbound campaigns, content strategies, and network-building operations through MCP. The common thread: they describe what they want in natural language, and the AI agent handles the execution using Crispy as its LinkedIn toolkit.