How to use AI code editors to manage sales leads

Connect Cursor or Claude Code to a lead tool through MCP, find prospects, check replies, and draft LinkedIn responses without leaving your editor.

Get started

Short answer

You can manage sales leads from Cursor, Claude Code, or another MCP client by connecting it to a lead tool through a Model Context Protocol server or REST API. The editor can then search for prospects, filter leads, check replies, and draft answers from plain English prompts. You still need to review the output and stay inside your account limits before messages go out.

What an AI code editor can do for lead work

An AI code editor is a chat window next to your files. When it can call a lead tool, it can run the same search you would run in a browser, then put the result in a table you can sort. You describe the customer out loud: founders at agencies with 20 to 50 people, heads of marketing at software companies that raised a seed round, or people who commented on a competitor's post. The editor turns that into a request.

Most people start with a narrow job. Find a list of 25 leads that match a profile. Remove anyone who works at a current customer. Show me the last ten replies from the campaign that targets agencies. Draft an answer to the one who asked about pricing. Keep the prompt small and the output checkable.

A useful first test is a list you can check by hand. Pick five people you already know fit, then ask the editor to find ten more like them and explain each match. If the reasons are weak, the prompt is weak. If the reasons are specific, save the search and reuse it.

The editor is not a replacement for judgment. It is a faster way to pull the list, see the context, and write the first draft. You still open the profile, read the reply, and decide whether the person is worth your time.

Browser workflow

  • Manual searching and clicking through individual profiles
  • Copying data into a spreadsheet by hand
  • Switching tabs to check whether a prospect replied

Editor workflow

  • One prompt to pull a list of leads that match a profile
  • Replies grouped by intent with the original message attached
  • A draft answer you can edit before it sends

Connect your editor to a lead tool through MCP

MCP is a way for an AI client to see tools a server exposes. You add the server once, then the editor can ask it for leads, inbox items, campaign actions, or whatever the tool supports. The setup looks different in each client, so start with the docs for Cursor, Claude Code, or the app you already use.

Do not connect every tool you can find. Pick the one that holds your leads and replies. A single connection is easier to test than a stack of integrations you will forget about.

If you use a local server, keep the config file in your project or home directory and note the command you used. When the connection breaks after an update, you will want the exact line to restore.

  1. 1

    Pick the client you already use

    If you live in Cursor, use Cursor. If you prefer a terminal, use Claude Code or Codex. The point is to avoid adding another place to check.

  2. 2

    Add the server config

    Open the MCP settings and add the server command or URL from the provider. For a remote server you paste a URL and finish OAuth. For a local server you add a command and any environment variables.

  3. 3

    Restart and list tools

    Ask the editor to list the tools it can see. You want names like search leads, list replies, draft reply, or create lead group. If the list is empty, the client did not load the server.

  4. 4

    Run a read-only test

    Ask for one lead or one inbox item before you ask it to draft anything. A read-only test tells you the connection works without touching outreach.

Find leads with plain english prompts

You do not need boolean strings. Give the editor the profile you want and the limit you can handle. A prompt that names the role, company type, location, and one buying signal will return a cleaner list than a prompt that says 'find good leads'. Use the output as a draft. Open a few profiles and check that the person matches the reason the editor gives.

When the table looks right, ask the editor to save those people to a lead group. Keep the group name tied to the search: 'uk-b2b-marketing-hiring-signal'. If the tool can score leads, set the fit line before you export or message. A list of 200 names is not useful if you cannot say why each one is there.

Do not ask for 500 leads on the first pass. Ask for 20, read the table, then ask for 20 more with the same profile. You will spot patterns in the weak matches faster than you will in a long list you never open.

Find a first list

Task
Find 20 heads of marketing at B2B software companies in the UK that match this profile and explain each match.
Prompt to use
Open five profiles and confirm title and company.

Tighten a broad list

Task
Remove anyone under 20 employees and anyone who is not in a marketing leadership role.
Prompt to use
The count should drop, not stay the same.

Save a working group

Task
Create a lead group called uk-marketing-hiring and add the remaining people.
Prompt to use
Group name includes the search reason.

Paste this job

Use the lead tool to find 20 people who match this profile: heads of marketing at B2B software companies with 20 to 100 employees in the UK. Only include people who posted about hiring an SDR or building an outbound team in the last 60 days. Put the results in a table with name, company, title, the post URL, and one sentence on why they fit. Do not create a campaign yet.

Check replies and decide who to answer

The inbox is where the editor earns its place. Instead of scrolling every thread, ask for the replies that need a human. You can group by intent: asked for a meeting, asked a question about price, referred you to someone else, or said no. The editor can pull the post or campaign that started the thread so you have context in the same view.

Use a short prompt. Long prompts with many conditions produce answers you have to rewrite. Start with one question, read the list, then ask for a draft for one person. If the tool can send replies on its own, set it to hand the conversation to you at the first reply until you trust the output. Otherwise keep the AI in draft mode and send the message yourself.

Example

  1. You

    List replies from the Northwind campaign in the last 48 hours. Put interested replies first and include the original post or message.

  2. Suggested reply

    Three replies. Dana at Acme asked for a demo next week. Priya at a 40-person agency asked whether the tool works with a small team. Mark said he is not the right person and pointed to his ops lead.

  3. You

    Draft an answer to Dana. Keep it under 80 words, suggest Tuesday or Thursday, and mention the scheduling link.

  4. Suggested reply

    Hi Dana, Tuesday or Thursday works on my side. You can pick a slot at the link below. If neither day works, send me two times that do and I will make one fit.

Read every AI draft before it becomes a habit. A reply that sounds like you will get a better response than one that sounds like a template, and you are the one who has to continue the conversation.

Keep outreach inside account limits

An editor can send prompts faster than a person can click, so add a brake. Decide how many invites and messages you will send each day, then set those limits in the lead tool, not in the prompt. The prompt should never say 'send to everyone'. It should say 'prepare a list' or 'draft for review'.

Use one campaign per message angle. If you test a new opening line for agency owners, keep it separate from the software founder campaign. That way a reply tells you which angle worked and you can pause the weak one without touching the other.

Write the limit down where you can see it. A note in the campaign, a line in your prompt, or a field in the tool is enough. The point is to stop a late-night session from turning into a week of cleanup.

Daily limit

Check
The tool stops at the same number every day, including follow-ups.
What good looks like
A prompt that asks for 'as many as possible' or a limit you raise after a slow day.

Send window

Check
Messages go out in the lead's working hours, not yours.
What good looks like
A campaign that sends at 2 a.m. because the editor was open late.

Pending invites

Check
Invites are withdrawn when they sit too long, and you review the acceptance rate.
What good looks like
Hundreds of pending invites with no campaign owner.

Reply handling

Check
You know whether the AI hands off, replies until interest, or books a meeting.
What good looks like
A campaign that sends on its own while you are still testing the copy.

Doing this with Kinlead

Kinlead has a REST Agent API and an MCP server, so Claude, ChatGPT, Cursor, Claude Code, Codex and other MCP clients can find leads, check replies and draft answers. Chat apps connect with OAuth; scripts use API keys. The connection to LinkedIn runs through Unipile, a third-party provider, and Kinlead never stores the LinkedIn password.

Pro costs $49 a month after a 7-day free trial. A card is needed to start the trial, nothing is charged that day, and it can be cancelled any time in Settings. Pro includes 1 user, 1 LinkedIn account, unlimited AI agents, leads and campaigns, and API access. Extra LinkedIn accounts cost $40 a month each, less per account in larger numbers. Enterprise is custom priced with unlimited users and accounts, SSO, dedicated onboarding and priority support. You can read the setup at Agent API and MCP.

Start with a read-only test: ask for one lead and one reply, then decide whether the output is worth acting on.

Frequently asked questions

Related

Get started with Kinlead

Prefer markdown? How to use AI code editors to manage sales leads.md