Let Claude or ChatGPT Research an Account and Verify Its Contacts (MCP)

To let Claude or ChatGPT research an account and verify its contacts, connect the SimpleVerifier MCP server, paste one staged prompt and read the table it returns. The assistant researches the company with its own web search, lists people in the roles you name, checks every address with SimpleVerifier and drafts short emails from cited facts. It stops before anything is sent.

This is for people who prepare outreach one account at a time. The template page is claude-chatgpt-account-research-outreach. About five minutes of setup.

Who does what

Keep this split clear, because it decides how much you trust each part:

• The assistant reads public pages, finds names and writes drafts. It can be wrong or out of date.
• SimpleVerifier checks whether an address or a domain can receive mail. It does not research companies or find contacts.
• You read the result and decide what to send.

What you need

• Claude or ChatGPT with custom connectors or MCP support, and web search turned on.
• A SimpleVerifier API key from the API keys page, or sign in with your SimpleVerifier account when the client asks.
• A plan with API and MCP access. See pricing.

Step 1: connect the server

The server URL is https://simpleverifier.com/api/mcp.

• In claude.ai or ChatGPT, add a custom connector with that URL and sign in. The Claude setup and ChatGPT setup pages show the clicks.
• In Claude Code, run this one command, with your own key:

claude mcp add --transport http simpleverifier https://simpleverifier.com/api/mcp --header "Authorization: Bearer YOUR_SIMPLEVERIFIER_API_KEY"

The MCP docs list Cursor and other clients, and describe every tool. Other write-ups cover Claude Code and Cursor and the ChatGPT connector.

Step 2: check the connection

Ask the assistant to run get_usage. It should report your plan, what the plan includes and the rate limit. If it cannot, the connection is not working yet.

Step 3: paste the staged prompt

The full prompt is on the template page with a copy button. Fill its three bracketed lines: your offer in one sentence, the target company and website, and the roles you want. It runs five stages in order:

1. Research. Read the company's own site and public pages. Write what they sell, who they sell to and two facts you can cite, with a source link beside every fact.
2. Find contacts. List up to 8 people in your roles from public pages. Give an email only if it is published or you supplied it. Otherwise give a likely pattern and label it guess.
3. Check the domain. Call check_domain and report the mail setup and whether the domain is a catch-all.
4. Verify. Call verify_emails with every address, guesses included. Drop every do_not_send address and mark send_with_care addresses as needs my call.
5. Draft. For each remaining address write an email of four sentences or fewer, using only the stage 1 facts and naming the link for each.

It ends with a table: contact, role, email, status, action and draft. Then the assistant lists what it was unsure about.

What the tools do

• check_domain returns a domain's MX, SPF and DMARC records, its disposable status and an optional catch-all probe.
• verify_emails checks up to 100 addresses in one call and removes duplicates. It waits up to 45 seconds for the list to finish and returns every result, or a job ID you finish with get_job_results, which returns 500 rows a page.
• verify_email checks one address.

Each result carries a status, an action and a reason_code. The API docs explain them. For a list of one account's contacts, 100 per call is plenty. For a list of thousands, upload a CSV instead.

Why guesses are verified too

An email pattern guess, such as first.last at the company domain, is only a guess. Running guesses through verify_emails is useful because a mail server that confirms the mailbox gives you real evidence. The check does not help on catch-all domains, where the server accepts any address. There the result is send_with_care, never valid. The catch-all study found that 39.1% of 4,943 sampled business domains accepted a made-up address, and 52.3% of Google Workspace domains did. So for guessed addresses at a catch-all domain, the honest answer is that you do not know.

Step 4: review the table

Read it as an editor:

• Open the cited link for each fact you plan to use. If you cannot find the fact on the page, cut it.
• Treat anything marked guess as unproven, even when it verified.
• Check that each contact still works at the company.
• Edit the tone. Short and plain beats clever.

Nothing has been sent. SimpleVerifier has no send tool, and the prompt tells the assistant to stop.

Where your judgment still matters

Drafts from an assistant can sound the same as everyone else's. Add one line that only you could write. And a verified address is not permission to email someone. Follow the rules that apply to your market, and honor opt-outs.

If you have many accounts, move to a batch flow instead: verify leads in n8n before Instantly, or clean a lead list before a campaign. For agent builders, see the AI agents page and the agent skill.

Frequently asked questions

Which MCP tools does the prompt use? check_domain and verify_emails, plus get_usage to check your plan.

Will it send the emails? No. It stops and shows a table.

Does SimpleVerifier find contacts? No. The assistant does, from public pages, and it can be wrong.