How to control LinkedIn outreach from an AI chat app

Connect an AI chat app such as ChatGPT or Claude to your outbound tool, then use plain prompts to find leads, check replies and draft answers.

Get started

Short answer

You connect your outbound tool to an AI chat app through an MCP server or API. Then you ask the AI to find leads, check replies and draft answers in plain language. You still set the campaign limits and decide which replies need your attention.

What chatting with your outreach tool actually does

A dashboard day goes like this: open the lead list, click a profile, check the group, open the inbox, then find your script in another tab. The work is scattered across screens. A chat connection puts the same actions behind a prompt.

You type what you want in the window where you already think and write. The AI calls the tool, gets the data back, and answers in a sentence. You do not need to learn a new dashboard or remember where a setting lives.

The connection does not remove the need for campaign rules. It changes how you read and act on the data. You still decide who to contact, what the message says and when a reply needs you.

Scattered dashboard work

  • Open the lead list and click through profiles to check fit
  • Open the inbox, read every reply and sort the serious ones by hand
  • Copy a lead's profile into a prompt, then paste the draft back into the tool

Chat request

  • Ask which leads in the group have the highest fit score and why
  • Ask which replies today mention pricing or a start date
  • Ask for a draft reply to one person, with the reason they became a lead

What you need before you connect anything

The setup has a few moving parts. You need a chat client that can call external tools, a connection layer from your outreach tool, and a credential that proves the app can read your workspace.

The chat client is the window you type in. The connection layer is an MCP server or REST API. MCP gives the AI a set of named tools, such as find leads or get replies. A REST API works the same way for scripts and custom clients.

The credential is either an OAuth login or an API key. Use OAuth when the chat app supports it. Use API keys for scripts or clients you run yourself.

Chat client

Piece
The window where you type requests
What it does
Claude, ChatGPT or Cursor

Connection layer

Piece
The protocol that exposes tools to the model
What it does
MCP server or REST API

Credential

Piece
Proof that the app can read your workspace
What it does
OAuth login or API key

Connect your AI client in three steps

Open the settings for the chat app you use. Find the section for connectors, MCP servers or plugins. The label changes by app, but the idea is the same: you are adding a tool the model can call.

Paste the server address from your outreach tool. If the app offers OAuth, use it. If you run a local MCP client, you may need an API key instead.

Then test with a read-only request. Do not start by asking the AI to add leads or change a campaign. Ask it to list something first, so you can see the connection working without side effects.

  1. 1

    Add the server

    Open your AI app's connector settings and paste the MCP server URL or API endpoint from your outreach tool.

  2. 2

    Authorize the connection

    Complete the OAuth login if it is offered. For a script or local client, paste an API key and keep it out of shared documents.

  3. 3

    Run a read-only test

    Ask the AI to list your lead groups or read the last few replies. Check that the names and campaigns match what you see in the tool.

Start with a read-only request, such as list my lead groups, before you ask the AI to add leads or draft messages.

Find leads from a prompt

Describe the buyer in the same words you would use to brief a new SDR. Include the job title, company type, location and the problem the person talks about. A vague prompt gives you vague leads.

Your outreach tool may score each lead against that profile. Ask for the score and the reason. Then you can see whether the AI understood the buyer or just matched a keyword.

Keep the first request small. Ask for 10 or 20 people, review them, then refine the prompt. A tight prompt you can reuse beats a long prompt you never check.

  • Job title and seniority, such as founder, head of sales or VP of marketing
  • Company type and size, such as a 40 person agency or a B2B software company
  • Location and language, if the offer depends on either
  • A problem or trigger, such as hiring SDRs, opening a new office or launching a product

Paste this job

Find 20 people who match this profile: [job title] at [company type] in [location]. They should care about [problem]. Put them in the lead group [group name]. Show me the 5 with the highest fit score and one sentence on why each one fits. Do not draft messages yet.

Read replies and draft answers without leaving the chat

Ask for replies by intent, not by date. A request like show me interested replies from today gives you a short list. A request like show me every reply gives you work you did not need.

When you find a real reply, ask for a draft that uses the person's words. The AI can see the lead's profile, the campaign goal and the reply. You still read it before it goes out.

If you want to review every message, use a hand-off mode in the campaign. If the AI replies on its own, check the campaign rules and the send window first. A chat connection is not a substitute for those settings.

Example

  1. Prospect

    Saw your note about hiring SDRs. We're trying to get the first sequence live before Q3. Do you help with that?

  2. You

    Yes. Teams at your stage usually need the list, the first message and a reply routine that someone can own. Are you free Thursday for 15 minutes?

  3. Prospect

    Thursday at 2 works. Send the invite.

Doing this with Kinlead

Kinlead has a REST Agent API and an MCP server. You can connect Claude, ChatGPT, Cursor, Claude Code, Codex and other MCP clients to find leads, check replies and draft answers. Chat apps connect with OAuth. Scripts use API keys. The setup is covered at Agent API and MCP.

Leads live in lead groups. Any campaign can point at any group, and you can export leads to CSV or import a CSV of LinkedIn profile URLs. The Steal Customers agent watches competitor company pages and optional founder or employee profiles, reads comments on recent posts, and turns commenters who look like buyers into leads. Each lead keeps the post URL, what the post was about and what they commented.

Outreach runs from your own LinkedIn account through Unipile. Kinlead never stores your LinkedIn password. Each account sends at most one action every 5 minutes, inside daily invite and message limits you set and inside a send window in the lead's time zone. Replies from every campaign land in one inbox, classified by intent, with interested replies first.

Frequently asked questions

Related

Get started with Kinlead

Prefer markdown? How to control LinkedIn outreach from an AI chat app.md