# How to connect LinkedIn leads to Claude or ChatGPT
> Set up an MCP server or REST API so Claude or ChatGPT can read your LinkedIn leads, check replies and draft answers without manual copy-pasting.
- HTML: https://getkinlead.com/linkedin-api-mcp-prospecting-workflow
- Markdown: https://getkinlead.com/linkedin-api-mcp-prospecting-workflow.md

**Short answer:** You connect LinkedIn leads to Claude or ChatGPT through an MCP server, which lets the AI call your lead tool directly, or a REST API, which your own scripts call. Either way the AI reads the lead's profile, the post that made them a lead and the full message history, so it can check replies and draft answers without you pasting anything.

## Why an AI agent beats copy-pasting into a chat window

The usual start is a browser tab with LinkedIn on the left and ChatGPT on the right. You copy a reply, paste it in, get something decent back, paste it into LinkedIn, then move to the next person. Around lead twenty you can't remember who said what, and the model has no idea who any of them are.

An agent with API access reads from the same place your campaigns write to. It sees the job title, the post that made the person a lead, the comment they left and every message since. The draft gets better because the model stops guessing.

The second gain is sorting. Ask which replies mention a competitor, or which ones asked about price, and you get an answer built from records rather than your memory of the last two days.

**Pasting into a chat window**

- You retype the lead's name, title and company every time
- The model only knows the one message you pasted
- You keep a separate note of who you already answered
- Asking where you left off gets a guess

**Agent reading the lead record**

- Title, company and source post arrive with the request
- The full message history is in context
- You can ask which replies mention a competitor and get a real list
- Drafts come back already knowing what the person cares about

## MCP or REST API: which connection fits your setup

Pick MCP if a person is going to sit there typing questions. You add the server once in Claude Desktop, Claude Code, Cursor, Codex or a ChatGPT connector, and the client then sees tools like find leads and check replies as things it can call. No code, no HTTP calls to write.

Pick the REST API if a machine is doing the work. A nightly job that pulls new replies into a sheet, a Slack bot that posts interested replies to a channel, a cron that drafts answers before you open your laptop. You write the calls and you hold the keys.

Most people need one of these, not both. The mistake is building a script for something you'd rather just ask in a chat box.

| | Connection | Who it suits | What you write |
| --- | --- | --- | --- |
| MCP server | Claude, Cursor or ChatGPT users who want to ask in plain English | One config entry and an API key |
| REST API | Scheduled jobs, internal dashboards and Slack bots | HTTP calls and an API key |
| Chat app with OAuth | Web chat sessions that connect through a sign-in | A one-time authorisation, no key handling |

## Setting up an MCP server in Claude or Cursor

The whole job takes about ten minutes and most of it is finding the config file.

1. **Install a client that speaks MCP.** Claude Desktop, Claude Code, Cursor and Codex all do. Use whichever you already have open every day.
2. **Create an API key in your lead tool.** Make a key for this client alone so you can revoke it without breaking anything else. Keep it out of shared docs and screenshots.
3. **Add the server entry to the config file.** Each client has its own settings file. Paste the server URL and the key into the MCP section, then save.
4. **Restart and check the tool list.** Reopen the client. You should see the lead tools listed. If they aren't there, the config path is wrong nine times out of ten.
5. **Run a read-only test first.** Ask it to list your five newest leads. Confirm the names match what you see in the tool before you let it draft anything.

## Writing a prompt that pulls real context from the lead record

The gap between a generic draft and one worth sending is what you put in the prompt. Ask the agent to fetch the record first, then draft in a second step. One call that does both tends to skip the fetching.

Name the fields you want used: the post they commented on, the comment itself, their title, the last two messages. If the model has to fill a gap, the draft reads like a mail merge and the person can tell.

Give it a shape as well. Length, tone, one question at the end, and the phrases you never want to see.

```
Fetch the record for this lead, then draft a reply.

Use only these fields from the record:
- name, title, company
- the post that made them a lead, and their comment on it
- the last two messages in the thread

Their latest message: [paste reply]
Goal: a 15 minute call about [product]
Tone: direct, no pleasantries
Length: under 60 words, one question at the end

Do not use: hope you're well, circling back, quick question, bullet points.
If a field is missing, say so instead of inventing it.
```

## Handling replies without losing the thread

Replies are where AI drafts earn their keep and where they do damage. A reply that ignores what the person actually asked is worse than sending nothing at all.

Pull the reply and the message it answers together, then ask for two things: what this person wants, and a draft that answers it in the fewest words.

Set a rule about when the AI stops. Anything touching price, contract terms or a complaint should reach you before a draft goes anywhere.

- **them:** We already use something for this. What would switching actually involve?
- **draft:** Fair question. Most teams run both for a quarter on one campaign, so you're comparing real replies instead of a demo. Would 15 minutes be useful to see whether that fits how your team works?
- **you:** Fair question. Most teams run both for a quarter on one campaign, so you're comparing real replies. Want me to send the setup steps so you can see the work involved before we talk?

## Keeping the account safe while AI does the typing

The API isn't the risk. Volume is. LinkedIn doesn't publish exact invite or message limits, so nobody quoting a number is telling you something official. Work inside whatever limit your account has and stay under it.

Patterns get accounts flagged more than totals do. Fifty invites in ten minutes, identical opening lines, a two-week-old account sending hard on day one. Spread sends across a window in the lead's timezone and keep the pace slower than feels efficient.

The other risk is text. If every message opens the same way, the person notices before LinkedIn does. Let the AI reference the specific thing that made someone a lead, and vary how the first line starts.

> Draft with AI, send from your own account, and keep the pace slower than you think you need.

## Doing this with Kinlead

Kinlead has a REST Agent API and an MCP server, so Claude, ChatGPT, Cursor, Claude Code and Codex can find leads, check replies and draft answers. Chat apps connect with OAuth and scripts use API keys. Leads sit in lead groups and are scored 0 to 100 against the customer you describe, and any campaign can point at any group.

Replies from every campaign land in one inbox, classified by intent with interested ones first, and you get an email when someone replies. Each campaign decides how replies are handled: hand over at the first reply, let the AI reply until there's real interest, or let it reply until a meeting is booked with your scheduling link.

The setup details are on the [Agent API and MCP](/features/agent-api-and-mcp) page.

## Related

- [Agent API and MCP](https://getkinlead.com/features/agent-api-and-mcp.md)
- [MCP integration](https://getkinlead.com/integrations/mcp.md)
- [Unified inbox](https://getkinlead.com/features/unified-inbox.md)
- [Claude integration](https://getkinlead.com/integrations/claude.md)

## Common questions

**Do I need a Sales Navigator seat to run this?**

Not if your lead tool uses LinkedIn's standard search. Sales Navigator adds filters, but the connection to Claude or ChatGPT works the same either way.

**Can the AI send messages, or does it only draft them?**

It depends on the tool. Some only draft and leave sending to you. Others send from your own LinkedIn account and let you choose whether a human takes over at the first reply.

**Will connecting an API get my LinkedIn account restricted?**

No tool can promise that. What lowers the risk is sending from your own account, staying inside a daily cap you set, and keeping the pace close to how a person actually works.

**What if my AI client doesn't support MCP?**

Use the REST API instead. Any language that can make an HTTP request works, and a short script that pulls replies into a sheet covers most of what MCP would give you.

[Create a Kinlead account](https://getkinlead.com/signup) or [book a call](https://getkinlead.com/demo) to talk through pricing and setup.
