Use it from Claude, Cursor, or any AI tool (MCP)

OpenLeads ships a Model Context Protocol server, so an AI assistant can find and verify leads for you mid-conversation:

"Find 10 dentists in Austin with emails we can actually send to, and draft a short intro for each."

It runs locally over stdio, with no extra dependencies and no keys. Searches go from your machine to the same public sources the CLI uses.

#Set it up

Install OpenLeads (pip install openleads, or pipx install openleads), then add it to your client.

Claude Code

bash

claude mcp add openleads -- openleads mcp

Claude Desktop (claude_desktop_config.json), Cursor (~/.cursor/mcp.json), Windsurf, Zed and others use the same shape:

json

{
  "mcpServers": {
    "openleads": { "command": "openleads", "args": ["mcp"] }
  }
}

No Python on the PATH? Use the npm wrapper instead: {"command": "npx", "args": ["-y", "openleads", "mcp"]}.

Tool	What it does
`find_leads`	Plain-English search ("fintech founders", "law firms in London"). Returns people and companies with an email, a deliverability tier (`safe`/`risky`) and evidence (`found`, `pattern`, `guessed`). Takes `count`, `verified_only` and `budget_seconds`.
`find_email`	A name plus a company domain gives the most likely address, with evidence.
`verify_email`	Checks one address: `safe`, `risky` or `bad`, and why.
`list_sources`	Lists what OpenLeads can search.

The model sees the same honesty labels you do:

found: the address was published on the web, or the mail server confirmed it.
pattern: built from a pattern we saw a real address use at that domain.
guessed: a common pattern with nothing confirming it. Don't send these without checking.

Tell your assistant to prefer tier: safe for anything it will email.