Sales Navigator + AI: collect first, then hand off to MCP

The current Linkydy MCP server does not accept Sales Navigator URLs or translate a prompt into Sales Navigator facets. The reliable workflow is hybrid: you choose precise filters in Sales Navigator, Linkydy collects the results into its CRM, and the agent works with that saved list.

Four things AI changes in a Sales Navigator workflow

1. You keep control of exact Sales Navigator filters

Choose title, geography, industry, headcount, school, past company, and other facets in Sales Navigator itself. This is more reliable than asking a model to guess LinkedIn's internal filter IDs. If you prefer prompt-driven search, use standard LinkedIn people search: apply exact filters there and pass its URL to search_by_url.

2. The extension creates the handoff

Collect the filtered Sales Navigator results with Linkydy. The same local lead storage is available to list_leads, so the agent can immediately review names, titles, companies, locations, profile URLs, and statuses. For work-email lookup, find_email takes one profile URL and requires your Apollo.io API key.

3. First lines written from context

Template fields like {FirstName}, {Company}, and {JobTitle} keep messages personalized at the mail-merge level; an AI agent goes further and drafts opening lines per prospect from what it knows about their role and company. You approve the angle once, and the agent applies it lead by lead.

4. Campaign status stays visible

The agent can call get_campaign_stats for sent, accepted, pending, and acceptance rate, and get_safety_status for today's invite count. The current MCP toolset sends connection requests but does not expose a general follow-up-message scheduler.

How it works under the hood

Linkydy ships its own MCP (Model Context Protocol) server. The implemented toolset is: get_account, search_leads, search_by_url, list_leads, find_email, send_connection, react_to_posts, comment_on_post, get_campaign_stats, and get_safety_status. Actions execute through the extension in your own browser session.

Set it up in four steps

Works with Claude Desktop and Claude Code, ChatGPT, Grok, Cursor — any MCP-compatible client. Client-specific walkthroughs live at Linkydy + Claude, Linkydy + ChatGPT, and the main MCP setup page.

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

Step four: back in the dashboard, toggle the MCP connection On. Keep your browser open while the agent works — Linkydy runs in your session, not in a cloud. That's the whole setup; the MCP page has troubleshooting notes per client.

Prompts to try on day one

• "I collected the Sales Navigator results in Linkydy. Call list_leads with filter ‘fintech’ and show names, titles, companies, locations, and profile URLs."
• "For this approved profile URL, call find_email. If it says not_configured, remind me to add my Apollo.io API key in Linkydy."
• "Use search_by_url with this standard LinkedIn people-search URL: [paste URL]. Collect 15 profiles, then call list_leads."
• "React to the two most recent posts from this approved profile. Draft any comment first and wait for approval."
• "Call get_safety_status, show today's invite count, and wait for my approval before one send_connection call."

Notice what these have in common: every prompt maps to an implemented tool and includes a review point. When you want raw data, export it manually from Linkydy using CSV export; CSV export is not currently an MCP tool.

Honest limits

• Your browser must be open. The agent orchestrates, but the extension executes in your live session. No browser, no actions — that constraint is also the safety feature.
• No artificial MCP cap. The agent should call get_safety_status, pace sends, and stop when LinkedIn signals its own limit; the extension auto-pauses on invite-limit and rate-limit responses.
• Premium feature. Lead collection and enrichment start free, but driving Linkydy from an AI agent requires a Premium plan.
• Review beats autopilot. The best results come from agent-drafted, human-approved outreach — approve the angle and the list, then let the pacing do its work.

FAQ