Credits, API and agents

How to connect people search to an AI agent with MCP

To connect people search to an AI agent, add Invenitor's remote MCP server at https://api.invenitor.com/mcp to your client with your key as a bearer token, then ask the agent to check your balance.

By Invenitor team. Updated .

Why connect a data tool at all?

Ask a language model for someone's email and it may give you one that looks right and is not real. The model is good at producing likely text, and an address follows a likely pattern. A tool call changes this: the agent asks a system and gets back what was actually found, or nothing.

MCP is the open standard that lets AI clients call tools like this. The project describes it as a USB-C port for AI applications (MCP introduction). Once it is connected, you can say "find me ten founders of climate startups in India" in your agent, and it uses the same search and credits as the app.

How do you connect it?

Remote MCP is on by default on every account. You need your key, from the API key page. In Cursor, add this to your MCP settings. Other clients use the same URL and header.

{
  "mcpServers": {
    "invenitor": {
      "url": "https://api.invenitor.com/mcp",
      "headers": {
        "Authorization": "Bearer pf_live_YOUR_KEY"
      }
    }
  }
}

For Claude Code, run this one line: claude mcp add --transport http invenitor https://api.invenitor.com/mcp --header "Authorization: Bearer pf_live_YOUR_KEY". The MCP setup notes also cover ChatGPT, VS Code, Windsurf and Codex. Never put your key in a URL, and keep the config file out of any shared repository.

What can the agent do once connected?

The server offers tools for each step. The main ones:

ToolWhat it doesSpends
get_balanceShows your remaining creditsNothing
people_searchRuns a plain-words search for up to 100 people20 credits (₹40)
get_searchReads the people from a searchNothing
open_contactGets the email and phone of someone from a search1 credit (₹2), only when found
find_contactLooks up one person from a LinkedIn or GitHub link1 credit (₹2), only when found

There are free tools for your contacts and shortlists too. Ask your agent to list what it can see once you have connected. Tool names can change as the server grows, so trust the list your client shows you over any page, including this one. A search stays in your account for 30 days, so an agent can return to it later without paying again.

How do you keep your agent from overspending?

An agent can spend credits quickly if you give it loose instructions. Put limits in your prompt:

  1. Tell it to call get_balance first and report the result.
  2. Set a budget in plain words: "Run one search. Open at most five contacts."
  3. Have it show you the shortlist and ask before it opens anyone.
  4. If your client supports approval prompts for tools, turn them on for the tools that spend.
  5. Tell it to use only addresses the tool returned, and never to guess one.

The server's own instructions ask agents to confirm with the person before opening a contact, but your client decides how strictly that is followed. Check your first few runs yourself.

What does a good prompt look like?

Here is one you can adapt. It sets the task, the budget and the checks in one place:

Use Invenitor. First check my balance.
Search for: operations heads at cold-chain logistics companies in Maharashtra.
Show me the 15 best matches with role and company, and explain why each fits.
Wait for me to pick. Then open only the people I pick, at most 5.
Do not guess any address. If a contact comes back empty, say so.
Draft a short note to each person for me to review. Do not send anything.

The advice in how to write a query applies here too. Give the agent the sentence you would say to a colleague, and let it show you the results.

What should an agent not do with this?

Do not connect a people search and an email-sending tool, then let the agent run unattended. An agent acting without review can message the wrong person, invent a detail about them, or keep going when it should stop. Invenitor does not send messages. Its Terms prohibit automated outreach and unconsented electronic marketing using its data, and they make you responsible for what you send.

A good pattern is that the agent finds and drafts while a person reviews and sends. Read each draft. Check any fact in it against the person's own page, since a model can still invent a detail even when the contact is real. See how to write a first email.

Connecting is free, and MCP uses the same credits as the app. If you are building your own product rather than using a client, read the REST API guide. Compare how other tools handle MCP on the comparison pages.

Common questions

Does MCP cost extra?
No. It is on every account and uses the same credits as the app and the API.
Can the agent send emails through Invenitor?
No. Invenitor does not send messages for you.

Keep reading