Linkydy

Product guide · Updated July 20, 2026

Claude, ChatGPT, Cursor, Gemini CLI, GitHub Copilot, Grok, Windsurf, and other MCP clients can use Linkydy as a focused prospecting toolset. This guide shows exactly what is exposed, which connection type each client needs, and how to run a first workflow.

What is MCP?

MCP (Model Context Protocol) is an open standard that gives AI assistants a consistent way to discover and call external tools. Instead of copying data between an AI chat and your browser, you register an MCP server with your client. The model can then request those tools as part of a conversation, subject to the approvals and controls in your client.

A LinkedIn MCP server, then, is a program that exposes LinkedIn actions — searching people, saving and listing leads, finding work emails, sending connection requests, reacting, commenting, and reading campaign status — 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.

Choose your AI client

The server is the same; only the client UI changes. Local clients launch linkydy-mcp with npx and pass the same command, arguments, and token. The current Linkydy extension guide also supports ChatGPT Desktop (Plus/Pro) through its local MCP server form. Hosted-only surfaces may still require a reachable HTTPS bridge.

Claude Desktop & CodeLocal stdio · setup guide**→** CursorLocal stdio · setup guide**→**

Gemini CLILocal stdio · shared config

GitHub Copilot in VS CodeLocal stdio · mcp.json

WindsurfLocal stdio · shared config

AIChatGPT DesktopLocal command form · setup guide**→** GGrokRemote HTTPS · setup guide**→**

+Other MCP clientsstdio or HTTPS

Where each part runs

Linkydy does not put your LinkedIn password or session cookie in the MCP configuration. The Chrome extension executes LinkedIn actions in the browser profile where you are already signed in; the local server relays tool requests to it over a local WebSocket.

Client typeConnectionWhat leaves the browser

Local clientsClaude, Cursor, Gemini CLI, Copilot, Windsurf, ChatGPT Desktopstdio launched with npxTool requests and returned lead data; never your LinkedIn password or session cookie

Hosted-only clientsSome web chat products and connector surfacesHTTPS optional bridge requiredTool requests and returned lead data pass through the endpoint you configure

Practical trade-off: Chrome must be open and the Linkydy connection must be enabled for tools to run. That keeps execution tied to the browser session you can see and stop.

Tools your agent gets

Tool	What the agent can do	Guardrail
get_account	Confirm Linkydy sign-in, Premium status, remaining free credits, and LinkedIn profile name	Metadata only
search_leads	Text search by keywords and title, optionally limited to 1st/2nd/3rd-degree connections; collects 1–100 leads	Search also saves matches
search_by_url	Collect from a LinkedIn people-search URL after you apply exact company, location, industry, school, or other filters in LinkedIn	LinkedIn people-search URLs only
list_leads	List saved leads with optional name, company, or title filtering	Reads local Linkydy data
find_email	Find a work email for one LinkedIn profile URL	Your Apollo.io API key required
send_connection	Send one optional personalized invite using {{FIRST_NAME}}, {{LAST_NAME}}, {{POSITION}}, and {{COMPANY}}	Message under 200 characters
react_to_posts / comment_on_post	React to 1–5 recent posts or comment on 1–3 recent posts for a profile	Explicit profile URL + bounded count
get_campaign_stats	Return sent, accepted, pending, and acceptance rate	Read-only
get_safety_status	Report invites sent today so the agent can self-pace	No artificial daily cap

Every write action still passes through the extension, but the MCP implementation does not impose an artificial daily cap. get_safety_status reports today's count so the agent can spread sends out, and the extension auto-pauses if LinkedIn responds with its own invite-limit or rate-limit signal. See LinkedIn's connection request limits for the broader risk picture.

Setup for local MCP clients

1. Install Node.js 18+ and the extension. Add Linkydy to Chrome, sign in, and keep Chrome open.
2. Copy your private pairing token. Open Linkydy dashboard → AI Agent (MCP). The token and On switch require Premium. Regenerating the token immediately disconnects clients using the old value.
3. Add the server to your MCP client config:

Canonical config generated by LinkydyCopy 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.

This is the exact config shape generated by the extension. Claude Desktop, Cursor, and Gemini CLI accept the mcpServers form above. Only the client wrapper changes: VS Code uses a top-level servers object in .vscode/mcp.json, while ChatGPT Desktop asks for the same Command, Arguments, and Environment values in a form.

Local bridge: linkydy-mcp listens on 127.0.0.1:8765. The current Chrome extension expects that exact loopback port, while the MCP client communicates with the server over stdio. If the port is busy, stop the conflicting local process and restart your AI client.

Prompt examples your agent can act on

Useful prompts specify the target, list size, desired action, and any point where you want a review before execution.

Text search

“Call get_account, then use search_leads with title ‘Head of Sales’, keywords ‘B2B SaaS Berlin’, and count 25. Call list_leads and show me the result before outreach.”

Exact filters

“Use search_by_url with this filtered LinkedIn people-search URL: [paste URL]. Collect 20 profiles, then list their names, titles, and companies.”

“For this approved profile URL, call find_email and report the result. If it says not_configured, remind me to add my Apollo.io API key in Linkydy.”

Outreach

“Call get_safety_status first. Draft an invite under 200 characters with {{FIRST_NAME}} and {{COMPANY}}, show it to me, then send one approved connection request.”

For exact facets, apply them in LinkedIn's standard people-search UI and pass its linkedin.com/search/results/people/… URL to search_by_url. The current MCP implementation does not accept a Sales Navigator URL directly.

The safety model

• No artificial MCP cap — the agent should call get_safety_status, show today's invite count, and spread sends out instead of bursting.
• LinkedIn-triggered auto-pause — the extension stops when LinkedIn returns its own invite-limit or rate-limit signal.
• 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 — free users can read the setup guide; Premium unlocks the pairing token and On switch. get_account is the only ungated tool exposed to the AI client.
• Private local token — keep LINKYDY_TOKEN out of repositories. Regenerate it in the extension to revoke old clients.

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

Prefer clicking over prompting? The same search, enrichment, and outreach controls are available directly in Linkydy's browser interface.