How to use the BounceHound MCP server in Claude, ChatGPT and Cursor
· The BounceHound team · 5 min read
Most email verification tools make you do the busywork: export a list, upload it, wait, download the results, and import them back. BounceHound's MCP server removes all of that. Connect it once, and your AI assistant can verify addresses and clean lists for you whenever a task involves email.
This guide explains what the MCP server does and walks through setup in Claude, ChatGPT and Cursor, step by step.
What is an MCP server?
The Model Context Protocol (MCP) is an open standard that lets AI assistants use outside tools. An MCP server is a service that offers those tools. Once you connect BounceHound's server, your assistant knows it can verify email addresses and decides on its own when to do it.
In practice, that means you can type something like this:
Clean this lead list before I import it into HubSpot.
Your assistant calls BounceHound, removes the addresses that would bounce, and reports back:
Verified 1,284 emails: 1,102 valid, 97 invalid, 61 catch-all, 14 disposable and 10 unknown. I kept the 1,102 valid addresses and flagged 3 typos, like jon@gmial.com, which should probably be gmail.com.
No exports, no uploads, no scripts.
The five tools your assistant gets
| Tool | What it does | Cost |
|---|---|---|
verify_email |
Checks one address and returns its status, the reason, and any typo suggestion | 1 credit |
verify_emails |
Checks up to 1,000 addresses and returns a summary plus each result | 1 credit per email |
clean_email_list |
Returns only the deliverable addresses from a list of up to 10,000, with counts of what was removed and why | 1 credit per email |
suggest_email_fix |
Catches typos like gmial.com or yahoo.con |
Free |
get_credit_balance |
Shows how many credits you have left | Free |
The verification tools are free for addresses that fail basic checks (bad syntax, no mail server, known disposable domains) and for results that come back "unknown". You only pay for real answers.
Every address gets one of five statuses: valid (safe to send), invalid (will bounce), risky (a catch-all domain that accepts everything), disposable (a throwaway inbox) or unknown (the mail server didn't answer).
Before you start: get an API key
All setups use the same API key.
- Create a free BounceHound account. You get 250 free credits, no card required.
- In the dashboard, open API & MCP and click Create key.
- Copy the key. It starts with
bh_live_and is shown in full only once, so store it somewhere safe.
The server address is:
https://bouncehound.com/api/mcp
In the examples below, replace YOUR_KEY with your own key.
Claude Code
Run this once in your terminal:
claude mcp add --transport http bouncehound https://bouncehound.com/api/mcp \
--header "Authorization: Bearer YOUR_KEY"
Start a new Claude Code session and run /mcp to confirm BounceHound is connected. Then just ask, for example: "Verify every email in contacts.csv and write the valid ones to contacts.clean.csv."
Claude Code is especially handy here because it can read and write files on your machine. It can open a CSV, verify the email column and save a cleaned copy in one go.
Claude Desktop
Open Settings → Developer → Edit Config and add BounceHound to claude_desktop_config.json:
{
"mcpServers": {
"bouncehound": {
"command": "npx",
"args": [
"mcp-remote", "https://bouncehound.com/api/mcp",
"--header", "Authorization: Bearer YOUR_KEY"
]
}
}
}
This uses mcp-remote to connect Claude Desktop to our hosted server, so you'll need Node.js installed. Save the file, restart Claude Desktop, and BounceHound's tools will show up in the tools menu of a new chat.
Claude on the web (custom connector)
On plans that support custom connectors, you can add BounceHound to Claude in your browser:
- Go to Settings → Connectors and choose Add custom connector.
- Name it BounceHound and use this URL:
https://bouncehound.com/api/mcp?api_key=YOUR_KEY
Web connectors can't send custom headers, so the key goes in the URL instead. Treat that URL like a password.
ChatGPT
ChatGPT connects to remote MCP servers through custom connectors in developer mode, on plans that include it.
- In ChatGPT, open Settings → Apps & Connectors → Advanced settings and turn on Developer mode.
- Go back to Apps & Connectors and choose Create.
- Name it BounceHound and add a short description such as "Verify and clean email addresses".
- For the MCP server URL, enter:
https://bouncehound.com/api/mcp?api_key=YOUR_KEY
- Set authentication to No authentication (your key is already in the URL), then create the connector.
- In a new chat, enable BounceHound from the tools menu and ask away: "Which of these emails are safe to send to?"
ChatGPT's menus change from time to time, so if a label looks different, look for the developer mode and connector settings. As with the Claude web connector, keep the URL private because it contains your key.
Cursor
Open Cursor Settings → MCP and add a new server, or edit ~/.cursor/mcp.json directly:
{
"mcpServers": {
"bouncehound": {
"url": "https://bouncehound.com/api/mcp",
"headers": { "Authorization": "Bearer YOUR_KEY" }
}
}
}
Save, and BounceHound should appear with a green status in the MCP settings. In Agent mode, Cursor can now verify addresses while it works. That's useful for checking seed data, cleaning a CSV in your repo, or testing the signup flow you're building.
Prompts to try
Once you're connected, plain English is enough. Your assistant picks the right tool.
- "Verify jane@acme.com."
- "Clean the emails in leads.csv and give me only the deliverable ones."
- "Check these signups for typos and fake addresses, then tell me which ones to email."
- "Remove disposable and catch-all emails from this list before I import it."
- "Before you send the outreach emails, verify every recipient and skip anything risky."
- "How many BounceHound credits do I have left?"
Tips
- Ask for a balance check before big jobs. Your assistant can call
get_credit_balancefirst so a large list doesn't stop halfway. - Decide what to do with risky addresses. By default,
clean_email_listremoves catch-all addresses, which is the safest choice for your sender reputation. Ask your assistant to keep them if you'd rather send to them separately. - Use one key per tool. Create separate keys for Claude, ChatGPT and Cursor so you can revoke one without breaking the others.
- Keep your key private. Don't commit config files with your key to a shared repo. If a key leaks, revoke it in the dashboard and create a new one.
Troubleshooting
- "Unauthorized" errors: check that the key is copied in full, including the
bh_live_prefix, and hasn't been revoked. - The tools don't appear: restart the app or start a new chat after adding the server. Most clients only load MCP servers at startup.
- "Insufficient credits": top up from the Buy credits page. Nothing is charged when a request fails for lack of credits.
Prefer plain code?
Everything the MCP server does is also available through the REST API, with the same key and the same credits. See the MCP server page for a full overview, or sign up free and connect your assistant in under a minute.