Give your AI agent a B2B contact search tool in one line of config

Published 18 September 2026 · TAPAC

Most agents are good at writing outreach and bad at knowing who to send it to. That gap is not a prompting problem — it is a missing tool. The Model Context Protocol (MCP) is the standard way to hand an agent a real capability instead of asking it to guess. TAPAC ships an MCP server, so the "find and verify contacts" step becomes a tool call.

The install

npx -y @tapacapi/mcp

That is the whole local install: no clone, no build, no Python. For clients that read JSON config (Claude Desktop, Cursor, Codex, Windsurf), the entry is:

{
  "mcpServers": {
    "tapac": {
      "command": "npx",
      "args": ["-y", "@tapacapi/mcp"],
      "env": { "TAPAC_API_KEY": "your_key" }
    }
  }
}

If you prefer Python, uvx tapac-mcp works the same way. There is also a hosted endpoint, and a hosted SKILL.md for agents that load skills rather than servers. Full notes: tapacapi.com/mcp.html.

What the agent can call

Verification happens at request time, over SMTP, per address. That is the difference from a stored list: what the agent hands you is what resolved a moment ago.

What happens with no key

Nothing breaks. The server returns onboarding text pointing at /get-key instead of throwing an error into the conversation — so you can wire an agent up before you have a key, and the agent tells the user what to do. 100 searches are free and no card is required.

Why this matters with real numbers

A sane sequence

  1. Install the server (one line above).
  2. Ask the agent for contacts with tight criteria — narrow beats broad.
  3. Use the returned verification status; do not send to anything unverified.
  4. Compare with your own numbers using the free lead cost & bounce-risk calculator.

Keep reading

Get a key: tapacapi.com/get-key.

← Back to TAPAC