How to Give an AI SDR Agent Real B2B Data with the LeadOcean MCP Server

Why an AI SDR agent needs a data layer

Ask a chat assistant for “10 heads of operations at mid-size food manufacturers, with emails” and, without tools, it will produce a confident list that is partly out of date and partly made up. The email addresses are the worst part: they look plausible ([email protected]) and bounce.

The fix is to give the agent a tool that searches a real database and returns structured records. The Model Context Protocol (MCP) is the standard way to do that: the assistant sees a list of tools, decides which to call, and works with the JSON that comes back. LeadOcean is built for exactly this. It describes itself as a GTM data API with a hosted MCP server and says it covers 693M people profiles and 63M companies, unified from 20+ sources into one schema.

What you need

• An MCP client: Claude, ChatGPT, Claude Code, Cursor, VS Code, Windsurf, Codex, Gemini CLI or an n8n workflow.
• A LeadOcean account. The free plan includes 1,000 records (a one-time allowance, no card) with full API and MCP access, which is enough to test your own segments.
• A written ideal customer profile (ICP): titles, seniority, industry, company size and countries. Vague briefs waste records. Our ICP building guide walks through one.

Step 1: Connect the MCP server

The server URL is the same everywhere: https://api.leadocean.io/mcp. Sign-in is OAuth in your browser, so you never paste a key into a config file.

Claude (web and desktop)

1. Open Customize → Connectors, then + Add → Add custom connector.
2. Name it LeadOcean, paste the server URL and click Continue.
3. Leave request headers empty, add the connector and sign in when prompted.
4. In a chat, turn LeadOcean on from + → Connectors.

ChatGPT

Turn on developer mode in ChatGPT's settings, add a new connector with the server URL, and sign in with your LeadOcean account.

Claude Code

Add it as a remote HTTP server, then run /mcp inside Claude Code to sign in:

claude mcp add --transport http leadocean https://api.leadocean.io/mcp

Cursor

Add the server to your MCP settings (~/.cursor/mcp.json), then sign in when Cursor asks:

{
  "mcpServers": {
    "leadocean": { "url": "https://api.leadocean.io/mcp" }
  }
}

Step 2: Size the market for free

Before your agent pulls anyone, make it count. LeadOcean's count tools spend no records, so they are the cheapest way to check that your filters mean what you think they mean. LeadOcean's own suggested first prompt is a good template:

How many VP-level sales leaders in Germany can I email?
Count only. Do not pull anyone.

If the number is far bigger or smaller than you expected, adjust titles, seniority or company size and count again. Ask the agent to call list_filters or list_enum_values when it is unsure of an exact filter value (an industry name, a country). Both are free too.

Step 3: Pull a small sample

Now ask for 5–25 people. Search returns thin records (name, title, company, location) without contact details, and costs 1 record per person returned. Read the sample like a human SDR would: are these the right roles at the right kind of company? If not, fix the brief now, while it is cheap.

Step 4: Enrich only the people you choose

Once the sample looks right, have the agent fetch the full profile for the contacts you approve. get_lead takes a LinkedIn URL, email or person ID and can reveal contact details. find_phone returns the phone numbers held for one person. Each costs 1 record. Enriching only approved contacts, rather than everyone a search returns, is where most of the savings come from.

For account-based work, start from companies instead: count_companies to size the market, search_companies or get_company to build the account list, then people at those accounts.

Step 5: Verify and hand off

Every enriched email should be treated as a candidate until it is verified, whichever data source it came from. Run the list through an email verifier, drop invalid and risky addresses, and keep bounces under 2% to protect your sending domain. Then pass the clean list to your CRM or sequencer. For larger lists, export_leads starts an asynchronous CSV export (1 record per row), and get_export returns the download link when it is ready.

The 13 LeadOcean MCP tools at a glance

A brief you can paste into your agent

Agents follow written limits well. This brief keeps a research agent inside a budget and stops it from inventing data. Swap in your own ICP:

You are an SDR research agent. Use the LeadOcean tools.

Ideal customer: VP or Head of Operations at US food & beverage
manufacturers with 200-1,000 employees.

Rules:
1. Always run count_leads first and tell me the total. Do not pull
   anyone until I say "go".
2. Pull at most 25 people per run with search_leads.
3. Enrich only the people I approve, with get_lead.
4. Never guess or construct an email address. If a contact has no
   email, say so.
5. Stop after 25 records in total and report what you spent
   (get_account).

Output: a table with name, title, company, domain, LinkedIn URL,
work email and email status.

Keeping cost and speed under control

• Count before you search, search before you enrich. Free counts catch bad filters; thin search records are cheaper than full profiles.
• Cap every run. A number in the brief (“stop after 25 records”) is the simplest budget control there is.
• Know the error codes. A 429 means you hit the request rate; wait for the Retry-After header and continue. A 402 means the plan's records are used up.
• Pick the plan by throughput. LeadOcean paces requests per account: 5 a second on Free, 50 on Solo, 200 on Agency and 10,000 on Platform. Interactive agents are fine on the lower tiers; overnight enrichment jobs need more.

Common mistakes

• Letting the agent pull before it counts. One vague prompt can spend hundreds of records on the wrong people.
• Enriching everyone in a search. Approve a shortlist first.
• Sending without verification. Even good data goes stale as people change jobs.
• Ignoring the licence. Check which plan covers client work or embedding data in a product before you build. Our LeadOcean review compares the plans.
• Skipping compliance. Outreach still has to follow GDPR, CAN-SPAM and local rules where your contacts live.

FAQ

What is the LeadOcean MCP server URL?

https://api.leadocean.io/mcp. Add it as a remote MCP server and sign in through your browser; there is no API key to paste.

Which AI tools work with it?

LeadOcean lists Claude (web, desktop and mobile), ChatGPT (developer mode), Claude Code, Cursor, Codex, VS Code, Windsurf, Gemini CLI and n8n.

Do count queries cost credits?

No. The count, filter, account and export-status tools are free. Search costs 1 record per result; a profile, company or phone lookup costs 1 record.

Can the agent verify the emails it finds?

Treat every email as a candidate until it is verified. Run addresses through an email verifier before a campaign and keep bounces under 2%.