How to generate leads with ChatGPT (using an MCP plugin)

TL;DR

• Time: about 15 minutes, most of it the one-time plugin setup.
• Cost: 6 LeadOcean records. The count is free, five sample leads are 5, one enrichment is 1. The Free plan has 1,000 records, no card.
• Result: ChatGPT counts your market, pulls a sample with title and company, and enriches one person, all from a chat.

ChatGPT writes a good outreach email and a bad prospect list. Ask it for "50 VPs of Sales" with no data source and it makes names up. The fix is to connect a real database over MCP, so the model calls a search tool instead of guessing.

Prerequisites

• A free LeadOcean account. No card.
• ChatGPT on the web. OpenAI's guide to adding a custom MCP server, October 2026, says the web app is required and workspace policy applies. It does not list which plans can add one. On a Business or Enterprise workspace, ask your admin.
• Optional, for the REST version of step 2: an API key from the app, exported as $LEADOCEAN_API_KEY, and curl.

The MCP route needs no API key. It signs in with OAuth.

Step 1: Add the LeadOcean server as a plugin

Add one URL and sign in. The steps follow OpenAI's guide to connecting a plugin, October 2026.

1. Open chatgpt.com/plugins and select the plus button.
2. Choose Add custom MCP server and name it LeadOcean.
3. Under Connection, paste the server URL. Set authentication to OAuth and leave client credentials empty.
4. Accept the risk warning and select Create as a plugin.
5. Install the plugin, then sign in to LeadOcean in the window that opens.

Server URL:  https://api.leadocean.io/mcp
Auth:        OAuth 2.1 (dynamic client registration, scope: mcp)
Tools found: 13

You should see 13 tools. Seven spend no records. The full list with record costs is on the ChatGPT setup page.

Step 2: Count your market before you pull anyone

Start with a count, because counting is free and tells you whether the audience exists. In a new chat, type @, pick the LeadOcean plugin, and send this prompt:

Use the LeadOcean plugin only. Call leadocean_count_leads for
jobLevel VP, jobFunction "Sales & Business Development", country US.
Count only. Do not pull anyone.

Naming the plugin and the tool keeps ChatGPT from browsing the web instead. That follows OpenAI's own advice to say which app to use.

The REST equivalent is one call. Send count=true as a query parameter, not in the body, with limit 1:

curl -s -X POST "https://api.leadocean.io/v1/people/search?count=true" \
  -H "x-api-key: $LEADOCEAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jobLevel": ["VP"],
    "jobFunction": ["Sales & Business Development"],
    "country": ["US"],
    "limit": 1
  }'

Both cost 0 records. Enum filters are exact strings, so if ChatGPT gets one wrong, ask it to call leadocean_list_enum_values first.

Step 3: Narrow until the number is a list you would work

A count of 129,494 is a market, not a list. Narrow it in the chat, one filter at a time, and re-count after each change.

Call leadocean_list_filters and find the company size filter.
Re-count with the same filters, limited to 51-200 employees.
Then show the top 5 industries from the breakdown.
Still count only.

Each re-count is free. Stop when the number is something one rep can work in a quarter. For most outbound teams that is a few hundred to a few thousand people, not 129,494.

Keep the filters you settle on. You will reuse them in the next step and in any export.

Two habits save records. First, count before every pull, because a count is free and a pull is not. Second, read the defaults note in the response. The count tool counts only mailable addresses unless you tell it otherwise, so a number with no email filter is larger. Label which one you quote in your own plan.

Step 4: Pull a sample of five

Now spend records, on purpose and in small amounts. Ask for five people so you can check quality before you scale.

Using the same filters, call leadocean_search_leads with limit 5.
Return a table: title, company, company size, country.
Do not reveal emails yet.

leadocean_search_leads costs 1 record per person returned, so this is 5. The REST version is the same body as step 2 without count=true and with "limit": 5. A search body is flat JSON, filters are arrays, limit runs 1 to 100, and you page with cursor.

Read the five rows. If the titles are wrong, fix the filter in chat and rerun with limit 5 again. That costs 5 more records, not 500.

Step 5: Enrich one person

Pick one row and enrich it. Enrichment takes exactly one of linkedin_url, email, phone or person_id. There is no name plus domain lookup, so enrich by the person_id the search returned.

Call leadocean_get_lead for the first person in that table.
Reveal the email. Show email_status and the current role.

That is 1 record, and a lookup that finds nobody costs nothing. The REST version:

curl -s -X POST https://api.leadocean.io/v1/people/enrich \
  -H "x-api-key: $LEADOCEAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"person_id": "PERSON_ID_FROM_STEP_4", "reveal_email": true}'

Check email_status before you send anything. There is no refund for bounced emails.

Do not ask ChatGPT to "find me leads" with no filters. A vague prompt makes the model pick filters for you, and it picks wide. A prompt that names the title, level, country and a limit gives you a bounded, predictable spend.

Step 6: Take the list out of the chat

ChatGPT is for sizing, sampling and one-off lookups. It is not where you work 2,000 leads. Move the list to your sequencer or CRM.

Two routes. Ask ChatGPT to call leadocean_export_leads with your filters. Or open the Exports page in the LeadOcean app. There you set filters, pick columns, see the record price before you start, and download a CSV. Bulk CSV export is not on Free. It is included on Solo at $97 a month. An export takes up to 50,000 rows a request and 500,000 a day.

LeadOcean has no sequencer, dialer or CRM. The CSV goes into the tool you already use.

What you get

Here is the real count from step 2, run through the LeadOcean MCP server on 2026-10-21. It cost 0 records.

{
  "total": 129494,
  "totalIsExact": true,
  "totalAll": 350517,
  "distinctCompanies": 58121,
  "credits": 0,
  "defaults": "Counted only addresses worth sending to (emailStatus: verified or catch_all_valid or catch_all)",
  "facets": {
    "industry": [
      {"value": "Financial Services", "count": 12677},
      {"value": "Software Development", "count": 10144},
      {"value": "Banking", "count": 7813}
    ],
    "company_size": [
      {"value": "10001+", "count": 23661},
      {"value": "11-50", "count": 20836}
    ]
  }
}

I trimmed the facets for length. Note what the number is. 129,494 is VPs with a verified, catch-all-valid or catch-all email. Without that email filter, the same filters match 350,517 people (totalAll).

A record-level search response has this shape, from the API reference (placeholder values, not a real person):

{
  "success": true,
  "data": [
    {
      "profile_data": { "...": "title, company, location" },
      "contact_data": { "has_email": true, "email_status": "verified" }
    }
  ],
  "meta": { "count": 5, "limit": 5, "credits": 5, "nextCursor": "abc123" }
}

Search returns a thin projection with no contact points. Emails come from enrichment with reveal_email set to true.

Troubleshooting

FAQ