Claude for sales prospecting: prompts and setup

Many salespeople already use Claude or ChatGPT every day for writing and research, so it is natural to try Claude for sales prospecting as well. Part of it works out of the box: research, profiles and first drafts. The rest, such as verified contacts, sending from your mailbox and tracking replies, needs a connected tool. The standard for connecting such tools is called MCP.

This guide explains what the assistants can do on their own, what a connected tool adds, how to add one in Claude and ChatGPT, and a workflow with prompts you can copy. Everything about Claude and ChatGPT here comes from Anthropic’s and OpenAI’s own documentation as of 2 October 2026. Those products change often, so the links at the end are worth a look before you rely on a detail.

In short:

• On its own, Claude can research companies on the web with citations and write well. It cannot verify contacts or send from your mailbox.
• MCP lets Claude and ChatGPT call tools in other software, such as a lead database or an outreach platform.
• Keep a person in the loop for anything that sends, and connect only servers you trust.

Prospecting with Claude: what it does on its own

Without any connector, Claude is good at research and writing. With web search turned on, it searches the live web and, according to Anthropic’s help page on web search, every answer includes citations so you can check the sources yourself. When you give it a specific URL, it can also read that page in full.

That is enough for a lot of the early work:

• Describing your ideal client. Give Claude a few of your best past clients and ask what they have in common. Turn the answer into a written ideal client profile.
• Researching one company before a call or a message: what it sells, recent news, open roles, who leads which team.
• Drafting a first line or a whole message for one prospect, based on that research.
• Checking your own copy. Ask Claude to cut your email in half or to point out every sentence that is about you instead of the reader.

Claude has no contact database, so it cannot give you a verified email address for a buyer. It cannot send from your mailbox or post from your LinkedIn account. It does not keep a list of who unsubscribed or who you already wrote to last month. Web search also finds what is public, which for small companies is often a website and little else.

Assistant alone or connected tool

A quick way to decide what you need is to sort your tasks by where the data lives.

If most of the tasks you care about sit in the first column, you may not need any connector yet. Start with research and writing, and add a tool when the list building and sending become the slow part.

What a connected tool (MCP) adds

MCP, the Model Context Protocol, is how an assistant reaches beyond the chat window. Anthropic open-sourced it on 25 November 2024 as a standard for connecting AI assistants to the systems where data lives. The MCP documentation calls it an open-source standard for connecting AI applications to external systems and compares it to a USB-C port: one connector shape that many devices understand.

In practice, a software vendor runs an MCP server. Your assistant connects to it, sees a list of actions the server offers, and calls them when your request needs one. For prospecting, those actions might be a company search, a contact lookup, adding people to a list, or drafting a campaign, depending entirely on the vendor.

Two things follow from that:

1. The server decides what your assistant can do through MCP. Read what a vendor’s server actually offers before you assume it can do a step.
2. Several sales tools now have MCP servers. Apollo, for example, says its agentic system works where you work, naming Apollo, Claude and ChatGPT. Compare servers on what they let you do and how they handle approvals.

How to add a connector in Claude

Anthropic calls these remote MCP integrations custom connectors. According to Anthropic’s help center, they are available in Claude, Cowork and Claude Desktop on the Free, Pro, Max, Team and Enterprise plans, with Free users limited to one connector. Connectors you add also work in the mobile apps.

On an individual plan:

1. Open Customize, then Connectors.
2. Choose “+ Add”, then “Add custom connector”.
3. Paste the server URL the vendor gives you.
4. Sign in when asked and review the permissions you are granting.

On Team and Enterprise plans, an owner adds the connector in Organization settings, under Connectors.

Anthropic warns that custom connectors connect Claude to services Anthropic has not verified. Connect only servers you trust and read the permissions during sign-in. For HypeLead’s own server, the developers page lists the exact steps.

In Claude Code

If you work in the terminal, the Claude Code MCP docs show one command for a remote server:

claude mcp add --transport http <name> <server-url>

Then run /mcp inside Claude Code and follow the steps in your browser to sign in. The docs warn that servers which fetch outside content can expose you to prompt injection, so verify each server before you add it.

ChatGPT, Cursor and VS Code

ChatGPT connects to MCP servers through developer mode. OpenAI’s MCP guide describes the steps: open Settings, then Security and login, turn on developer mode, then add your server URL as a plugin. The same guide says ChatGPT currently requires manual confirmation in any conversation before write actions can be taken. OpenAI’s help center lists which plans include developer mode, so check it for your account.

Code editors such as Cursor and VS Code also support MCP, according to the MCP documentation. That is useful if your team builds its own prospecting scripts, but for day-to-day sales work a chat assistant is the more natural home.

A prospecting workflow that holds up

The workflow below works with Claude alone for the research steps and with a connected tool for the rest. The order matters more than the exact prompts.

1. Write down who you want. One paragraph on the industry, company size, country, the role you sell to and the problem you solve.
2. Find companies. With a connected tool, ask for a list that matches the profile. Without one, ask Claude to research a short list of named companies you already know.
3. Check fit before you write. Ask for a one-line reason per company. Drop anything where the reason is vague.
4. Research each remaining company briefly. Recent news, hiring, a new product, a new market.
5. Draft one message per person, built on one specific observation from the research.
6. Read every draft yourself. Edit, cut, and approve. Only then let the connected tool send, from your own mailbox.
7. Sort the replies, and write down who said no so they never get a second campaign.

Prompts that work well at each step:

Save the prompts that work in a shared document, so everyone on the team starts from the same wording and you can see which version produced the replies.

Two habits make the output better. Ask Claude to mark anything it could not verify, and keep the sources it cites next to each company. When a prospect replies “where did you get that?”, you will have the answer.

Safety: what to check before you connect a server

An MCP server can read and change data in another system on your behalf. Treat connecting one like giving a new colleague a login.

• Vendor. Connect only servers from vendors you would trust with the same data in their own app.
• Permissions. Read the scopes or permissions during sign-in. A prospecting tool rarely needs more than access to its own data.
• Approvals. Find out which actions run without asking. Anything that sends a message, spends money or deletes data should wait for your click.
• Outside content. Anthropic’s Claude Code docs warn that servers which fetch outside content can expose you to prompt injection: text on a web page that tries to give your assistant instructions. Be careful when one session combines web research with actions that send or delete.
• Regular review. Once a month, look at the list of connected servers and remove the ones nobody uses. Each server you keep is one more service that can see what you share in the chat, so keep the list short.
• Channel rules. An assistant does not change LinkedIn’s rules or email law. Our guide to safe LinkedIn limits covers what LinkedIn allows. Email rules differ by country. In Czechia (Act No. 480/2004 Coll.) and Germany (UWG, section 7), for example, commercial email may go only to people who gave prior consent or to an existing customer you write to about your own similar products or services, and that applies to company addresses too. Our guide to B2B cold email and GDPR goes through the rules country by country.

What an assistant still will not do for you

Even with a good server connected, a few parts stay with you.

Judgement about who deserves your time. An assistant ranks well against a profile you wrote. It cannot tell that a prospect is a poor fit for reasons you never put in writing.

Mailbox health. Warming up a mailbox, setting up SPF, DKIM and DMARC, and watching bounce rates belong to the sending tool, not to the chat. Our guide to choosing an outbound stack lists what to look for.

Data hygiene. If the same company sits in your CRM, a spreadsheet and the connected tool, an assistant will work with all three copies. Decide which system holds the record and point the assistant at that one.

The decision to send. Keep the last click for yourself, at least until you have read enough of the assistant’s drafts to trust its judgement. Our guide to what an AI SDR can’t do goes into how much to hand over.

Using HypeLead from Claude

HypeLead has a hosted MCP server that works in Claude, ChatGPT, Cursor and VS Code. You connect it by signing in with your HypeLead account through OAuth, the sign-in step from the connection guide above, so you never paste a password into the chat. The MCP server and the public REST API are part of every HypeLead plan.

Through it, you can tell your assistant what you want in your own words, and the AI sales agent does it in HypeLead: searching 400M+ contacts in our global search and Czech company registers, judging leads against your ideal client profile and writing personal messages. Anything that changes data, spends credits or sends a message asks first: you see a preview with the credit cost or the full message, and it runs only after you say yes in the chat. Campaigns go out from your own mailbox and LinkedIn, and replies are drafted by AI and, by default, sent on your click.

Setup steps and a sample command are on the developers page. The full flow is on the product page, and the use case for B2B services shows how a small service firm would use it.

---

Want your AI assistant to find clients and draft campaigns in HypeLead? Your next step is at the end of this page.