Linkydy

Claude, ChatGPT, Grok, and Cursor can already write your outreach messages. With Linkydy they can also send them — search LinkedIn and Sales Navigator, save prospects to a CRM, find verified emails, and queue connection requests, all by prompt. This page explains what an MCP server is, how Linkydy's works, and how to connect yours in about two minutes.

What is MCP?

MCP (Model Context Protocol) is an open standard, introduced in late 2024 and since adopted by every major AI client, that gives AI assistants a uniform way to discover and call external tools. Instead of copy-pasting between your AI chat and your browser, you register an "MCP server" with your client, and the model can then invoke that server's tools directly — with your approval — as part of a conversation.

A LinkedIn MCP server, then, is a program that exposes LinkedIn actions — searching people, saving leads, enriching contacts, sending connection requests — as tools an AI agent can call. Linkydy ships one as an npm package, linkydy-mcp, and pairs it with the Chrome extension that actually performs each action inside your logged-in LinkedIn session.

Architecture: nothing leaves your machine

Most "AI + LinkedIn" tools want your password or an exported session cookie so a cloud bot can log in as you. Linkydy's design is deliberately different — the chain runs entirely on your computer:

AI agent (Claude · ChatGPT · Grok · Cursor) │ Model Context Protocol ▼ Linkydy MCP server npx linkydy-mcp │ local WebSocket (your machine only) ▼ Linkydy Chrome extension enforces caps & pacing │ ▼ LinkedIn — inside YOUR logged-in browser session

The agent never touches LinkedIn directly, and no third party sits in the middle. The MCP server is a thin relay; the extension does the work in your own browser, from your own IP, with your session untouched. No password sharing, no cookie export, no datacenter logins — the three patterns LinkedIn flags hardest. The honest trade-off: your browser has to be open for actions to run. That is also why this architecture is safer than any 24/7 cloud bot.

Tools your agent gets

Tool	What the agent can do	Guardrail
Search leads	Run LinkedIn people search and Sales Navigator searches with filters: job title, company, location, industry, language, past role, school	Read-only
Save & list leads	Save matches to your Linkydy CRM, list what is already there, check statuses and notes	Your data, local storage
Enrich contacts	Run Find Email / Find Phone on saved leads; results land on the lead card	Uses credits
Send connection requests	Queue personalized invites with {FirstName}, {Company}, {JobTitle} fields	Daily cap + pacing
React & comment	React to prospects' posts and leave comments to warm up outreach	Daily cap + pacing

Every write action passes through the extension's limiter. The agent can ask for 500 invites; the extension will still send only what the daily cap allows, at a human-like pace. See LinkedIn's connection request limits for why that matters — LinkedIn itself caps invites at roughly 100–200 per week.

Setup: connected in about two minutes

1. Install the extension and sign in. Add Linkydy to Chrome, then sign in with your email. Free credits, no card.
2. Copy your pairing token. Open the Linkydy dashboard → AI Agent (MCP) → copy the pairing token.
3. Add the server to your MCP client config:

{ "mcpServers": { "linkydy": { "command": "npx", "args": ["-y", "linkydy-mcp"], "env": { "LINKYDY_TOKEN": "PASTE_YOUR_TOKEN_HERE" } } } }

1. Toggle the connection On in the dashboard's AI Agent (MCP) panel. Your client will list the Linkydy tools on its next start.

The exact config file location depends on your client — follow the guide for yours: Claude Desktop & Claude Code, ChatGPT, Grok, or Cursor.

What a session looks like

You: Find 20 heads of sales in Berlin and send each a personalized invite.

Agent: ✓ Searched LinkedIn — 20 matches ✓ Saved to Linkydy CRM ✓ Found 14 verified emails ✓ 20 invites queued at a safe pace

Behind that transcript: the agent called the search tool with title and location filters, saved the results, ran enrichment, then queued invites through the extension — which spaced them out with human-like delays. You watch it happen in your own browser and can stop it at any point. The same pipeline works from a Sales Navigator search; there is a dedicated walkthrough on the Sales Navigator MCP page.

The safety model

• Daily caps — enforced by the extension, not the agent. A prompt cannot raise them.
• Human-like pacing — randomized delays between actions instead of machine-gun bursts.
• Your session only — no password sharing, no cookie export, no remote logins. Actions come from your browser and your IP.
• Browser-open requirement — Linkydy is not a 24/7 cloud bot. If Chrome is closed, nothing runs. That looks like a limitation; it is the reason your activity pattern looks human.
• Premium gate — MCP access is a Premium feature. Free accounts get an explicit premium_required response, never a silent failure.

For the full risk picture — what LinkedIn actually flags and how to stay under the radar — read Is LinkedIn automation safe?

Pick your client

Claude

Claude Desktop config file + claude mcp add for Claude Code. Includes Sales Navigator prompts.

ChatGPT

Developer-mode connector setup and what ChatGPT can run through Linkydy.

Grok

Add Linkydy as a custom connector and prospect from Grok.

Cursor

.cursor/mcp.json setup — research and reach leads without leaving the editor.

Prospecting from Sales Navigator specifically? Start with the Sales Navigator MCP workflow. Prefer clicking over prompting? Everything the agent does is also available by hand in Linkydy's outreach automation.