BlitzAPI MCP server · one URL, live data in every chat

Claude MCP: how to connect Claude to live B2B data

The short answer

To set up Claude MCP for B2B data, add a remote MCP server as a custom connector: in Claude, open Customize, then Connectors, add https://api.blitz-api.ai/mcp and sign in with your BlitzAPI account. Claude can then search and enrich people and companies, find verified work emails and search job postings while you chat, and each tool call counts records like an API call.

Updated
Read
8 min
Clip
0:16
On this page14 sections
  1. Why can’t Claude look up live B2B data on its own?
  2. What is an MCP server?
  3. What is a Claude connector?
  4. How do you set up Claude MCP for B2B data?
  5. What can you ask Claude once it has live B2B data?
  6. What happens when Claude calls the MCP server?
  7. How many records does a question use?
  8. Can ChatGPT, Cursor and VS Code use the same MCP server?
  9. Is it safe to connect Claude to a data MCP server?
  10. How do you add the BlitzAPI MCP server to Claude Code?
  11. How do you add it to Cursor, VS Code or Claude Desktop?
  12. What did the real run send and receive?
  13. How should an agent handle large pulls?
  14. Sources and method
From the explainer BlitzAPI × ZeroClick: built for agents to buy, 0:14 to 0:30.

Ask Claude who runs sales at one of your target accounts and it answers from memory: the org chart as it was when its training data ends, and no work email at all. An MCP server changes that. Add one URL to Claude, sign in, and Claude can look up live companies, people, verified work emails and job postings while you chat.

This guide shows how, step by step, with the BlitzAPI MCP server. MCP is one of the ways AI agents use BlitzAPI, next to the API and Agent Skills; the agentic commerce guide covers the newest one, agents that buy the data on their own.

Why can’t Claude look up live B2B data on its own?

Because a model answers from what it learned in training, and org charts change every week. People change jobs, get promoted and leave: the VP of Sales a model remembers may be gone. And a verified work email is not something a model can recall. It has to be checked against the company’s mail server, today.

“Who runs sales at Northwind today, and what's their work email?”

Training data endsToday

What the assistant remembers

VP of SalesNorthwind, when its training data ends

Has left the company since

What is true today

Head of SalesJoined after the training data ends

Work email not in its memory

From memory, an assistant sees the org chart as it was when its training data ends. Who holds the role today, and their work email, are out of its reach.

A web search can turn up a name. It can’t give you a work email that has been proven to work, and it won’t build a list of 200 accounts for you. For that, the assistant needs a live data source it can call.

What is an MCP server?

An MCP server is the plug between an AI assistant and a live system. It tells the assistant which tools it offers and what each one does; the assistant calls them when a question needs them, and the answer comes back into the conversation.

Definition MCP server also called Model Context Protocol server
A connector that gives an AI assistant tools it can use: it tells the assistant what each tool does, and the assistant calls them during a conversation to read data or take an action.

PreciselyAn MCP server exposes tools, resources and prompts to AI applications over the Model Context Protocol, an open standard introduced by Anthropic in November 2024 and donated in December 2025 to the Agentic AI Foundation, a directed fund of the Linux Foundation. A remote MCP server is reached by URL and usually signs the user in with OAuth.

MCP stands for Model Context Protocol. It is an open standard, so one server works in several AI tools: Claude, Claude Code, Cursor and VS Code all take the same address. In the BlitzAPI MCP server, each tool runs one BlitzAPI endpoint, with the same parameters and the same answer as the API.

  1. YouAsk in plain words
  2. Your AI toolClaude, Cursor or VS Code picks the tool it needs
  3. BlitzAPI MCP serverapi.blitz-api.ai/mcp, signed in as you
  4. BlitzAPI APIOne tool runs one endpoint
  5. Live recordsPeople, companies, verified emails, jobs

The answer comes back into the chat

One URL connects your AI tool to the BlitzAPI MCP server: the tool picks what it needs, the server calls the matching endpoint with your account, and live records come back into the chat.Watch this moment 0:19

What is a Claude connector?

A connector is how Claude plugs into an outside service. Some come from Anthropic’s directory; a custom connector is one you add yourself, by pasting the URL of a remote MCP server.

Definition Claude connector also called custom connector
An outside tool or data source plugged into Claude, so Claude can read from it or act in it while you chat. A custom connector is one you add yourself by pasting the URL of a remote MCP server.

PreciselyIn Claude, connectors link the assistant to external services: some come from Anthropic's directory, and custom connectors are remote MCP servers added by URL. Custom connectors are available on the Free plan (one at a time), Pro, Max, Team and Enterprise; on Team and Enterprise, an owner adds them for the organization first.

Custom connectors work on every Claude plan: Free (one at a time), Pro, Max, Team and Enterprise. On Team and Enterprise, an owner adds the connector for the organization first, then each person connects it with their own account. Claude reaches the server from Anthropic’s cloud, so the server has to be public, which the BlitzAPI one is.

How do you set up Claude MCP for B2B data?

In three steps: get a BlitzAPI account, add the server as a custom connector, then switch it on in a chat.

  1. Get a BlitzAPI account. Every account starts with 1,000 free records, no card.
  2. Add the connector. In Claude, open Customize, then Connectors. Click Add, choose Add custom connector, name it BlitzAPI, paste https://api.blitz-api.ai/mcp and click Continue. Keep the detected sign-in settings, choose Sign in now, click Add, and log in with your BlitzAPI account in the window that opens. There is no API key to copy.
  3. Switch it on and ask. In a chat, click the + button, open Connectors and turn BlitzAPI on. Then ask in plain words.

Claude Code, Cursor, VS Code and Claude Desktop take the same URL; their setup is in the builders section at the end of this guide.

What can you ask Claude once it has live B2B data?

Anything a prospect list or an account brief needs, in plain words. A few questions to start with, and the BlitzAPI endpoints each one runs:

  • “Find 25 VPs of Marketing at B2B SaaS companies in the US with 50 to 200 employees. Show name, title, company and profile URL in a table.” Claude runs a People Search.
  • “Using BlitzAPI, find the Head of Sales at acme.com and get their work email.” Claude chains Domain to LinkedIn, Waterfall ICP and Email Enrichment.
  • “Which companies headquartered in Texas are hiring SDRs right now?” Claude turns job postings into an account list with TAM by Jobs.
  • “How many records do I have left on my BlitzAPI plan?” Claude checks your account, which uses no records.

What happens when Claude calls the MCP server?

Claude splits your question into tool calls, each one a call to the API, then writes the answer from what comes back. We ran the second question on October 16, 2026, with the BlitzAPI MCP server, on a real company instead of acme.com. The company and the person found are withheld.

Real run · October 16, 2026 · BlitzAPI MCP server

“Using BlitzAPI, find the Head of Sales at ‹company›.com and get their work email.”

  1. company_find_by_domainCompany page found from the domain1 record
  2. people_waterfall_icpA Chief Revenue Officer, found by the first title tier1 record
  3. people_find_work_emailNot Found: no address it could prove0 records

Used for the whole question2 records

A real run: three tool calls, two records. The email call found no address it could prove, so it returned Not Found and counted nothing.

The first call found the company page from the domain. The second searched that company with a cascade of sales titles (Head of Sales or Chief Revenue Officer first, then VP of Sales, then Sales Director) and found a Chief Revenue Officer at the first level: the method the decision makers guide explains. The third looked for a work email it could prove, found none, and said so.

When an address can’t be proven, the answer is Not Found, not a guess.

How many records does a question use?

As many as the items it returns. A record is one company, one person, one email address, one phone number or one job posting, and a call that finds nothing counts nothing. The run above used two: one company and one person.

Each tool result reports the records it used, and tool calls run on your BlitzAPI plan exactly like API calls. Checking your plan or reading the changelog uses no records. For a pull of thousands of rows, ask Claude for a script that calls the API rather than one giant tool call: the rows land in a file instead of filling the chat. Plans are on the BlitzAPI pricing page.

Check yourselfYou ask Claude, through the BlitzAPI MCP server, for 25 VPs of Marketing and their work emails. What is the most this question can use?
Show the answer key
  1. 25 records: the emails come with the people.A verified email is a record of its own, counted when it is found. A people search returns profiles, not emails.
  2. 50 records: 25 people and up to 25 emails.Each person returned is one record, and each email found is one more. An email that cannot be proven comes back Not Found and counts nothing, so 50 is the most, not the bill.
  3. Nothing: tool calls inside Claude are free.Your Claude plan pays for Claude. The data runs on your BlitzAPI plan: tool calls count records exactly like API calls.

Can ChatGPT, Cursor and VS Code use the same MCP server?

Cursor and VS Code, yes: they take the same URL and open the same sign-in. ChatGPT accepts custom MCP servers in Business, Enterprise and Edu workspaces, on the web: an admin or owner adds the server from Plugins, with Add custom MCP server, then people in the workspace select it with @ in a chat. The BlitzAPI setup guide covers Claude, Claude Code, Cursor, VS Code and Claude Desktop.

Is it safe to connect Claude to a data MCP server?

Yes, with the care any connection to your accounts deserves. Sign-in goes through OAuth: you log in on BlitzAPI’s page, and Claude never sees your password. Anthropic doesn’t review custom connectors, so connect only servers you trust; the official BlitzAPI server lives at api.blitz-api.ai/mcp, run by BlitzAPI. If you use an API key instead, treat it like a password and keep it out of any file you share or commit.

For builders

Everything above works from a chat. This part is for developers in Claude Code, Cursor or VS Code: the setup for each, the calls from the real run, and how to keep large jobs out of the chat.

How do you add the BlitzAPI MCP server to Claude Code?

One command, then sign in from Claude Code:

claude mcp add --transport http blitz-api https://api.blitz-api.ai/mcp

Run /mcp inside Claude Code, select blitz-api and sign in with your BlitzAPI account. To send an API key instead, pass it as a header:

claude mcp add --transport http blitz-api https://api.blitz-api.ai/mcp \
  --header "x-api-key: YOUR_API_KEY"

How do you add it to Cursor, VS Code or Claude Desktop?

In Cursor, open the MCP settings and add the server to mcp.json; Cursor then asks you to sign in:

{
  "mcpServers": {
    "blitz-api": {
      "url": "https://api.blitz-api.ai/mcp"
    }
  }
}

In VS Code, the file is .vscode/mcp.json:

{
  "servers": {
    "blitz-api": {
      "type": "http",
      "url": "https://api.blitz-api.ai/mcp"
    }
  }
}

Older versions of Claude Desktop don’t connect to remote servers on their own and go through the mcp-remote wrapper (npx -y mcp-remote https://api.blitz-api.ai/mcp). Recent versions use the connectors of your Claude account.

What did the real run send and receive?

The second call, as it ran (company withheld):

{
  "company_linkedin_url": "https://www.linkedin.com/company/<company>",
  "cascade": [
    {
      "include_title": ["Head of Sales", "[Chief Revenue Officer]",
                        "[CRO]"],
      "exclude_title": ["assistant", "intern"]
    },
    {
      "include_title": ["VP Sales", "Vice President Sales",
                        "Vice President of Sales"],
      "exclude_title": ["assistant", "intern"]
    },
    {
      "include_title": ["Sales Director", "Director of Sales"],
      "exclude_title": ["assistant", "intern"]
    }
  ],
  "max_results": 1
}

It came back with one person, icp: 1 and ranking: 1, a current job titled Chief Revenue Officer, and "records_used": 1 in its fair_usage block. The email call answered "found": false, "email": null and "records_used": 0. Each tool runs one endpoint of the API reference: this one is POST /v2/search/waterfall-icp-keyword.

How should an agent handle large pulls?

With a script, not a giant tool call. Tool results land in the model’s context, so ask for a Python or JavaScript script that pages through the results and saves them to a file. Search tools take max_results and a cursor you send back until it is null, and a search stops at 50,000 results for people, company and TAM searches, 10,000 for the employee finder and 5,000 for jobs: split bigger pulls into narrower filters. The MCP server page lists the 24 tools, and the setup guide covers troubleshooting.

Sources and method

Published
API calls run on

The real run was made on October 16, 2026 with the BlitzAPI MCP server, against a real company. The company and the person found are withheld; the job title and the records counts are kept as returned.

The Claude and ChatGPT menus were read in their help centers on October 16, 2026. Both products move their settings often: if a label differs, their help centers have the current path.

The first diagram is illustrative: Northwind and its people are fictional.

Questions this raises

Is MCP only for Claude?

No. MCP, the Model Context Protocol, is an open standard that Anthropic introduced in November 2024 and donated in December 2025 to the Agentic AI Foundation, a fund of the Linux Foundation. Claude, ChatGPT, Cursor, VS Code, Gemini and Microsoft Copilot support it, so one MCP server serves several AI tools at the same URL.

Does Claude Pro allow custom connectors?

Yes. Custom connectors, the remote MCP servers you add by URL, work on Claude Pro and Max, on Team and Enterprise, and on the Free plan with one custom connector at a time. On Team and Enterprise, an owner adds the connector for the organization first, then each member connects it with their own account.

Can ChatGPT connect to an MCP server?

Yes, in ChatGPT Business, Enterprise and Edu workspaces, on the web. An admin or owner adds the server from Plugins with Add custom MCP server, enters its public URL and sets up sign-in, and people in the workspace then select it with @ in a chat. OpenAI lists custom MCP servers as a web-only feature.

What is the difference between an MCP server and an API?

An API is a set of endpoints that code calls with exact parameters. An MCP server wraps those endpoints as tools that an AI assistant can discover, read about and call by itself during a conversation. In the BlitzAPI MCP server, each API tool maps to exactly one API endpoint, with the same parameters and the same response.

Does using Claude with an MCP server cost extra?

Claude costs what your Claude plan costs. The data source bills its own way: on BlitzAPI, tool calls run on your BlitzAPI plan and count records exactly like API calls, so each item returned counts and a call that finds nothing counts nothing. Every BlitzAPI account starts with 1,000 free records, no card.

Which B2B data providers offer an MCP server for Claude?

Several B2B data vendors now publish one. The BlitzAPI MCP server runs at https://api.blitz-api.ai/mcp, hosted by BlitzAPI, with sign-in through your BlitzAPI account and one tool per API endpoint for people, companies, work emails and jobs. Whichever you pick, prefer an official server run by the provider, with OAuth sign-in and documented tools.

Data shouldn't be
the problem

Grab. Connect. Build.