Reads are safe
Autonomy policy: all reads are agent-safe because they only return account state, cached markets, and prepaid local-lead intelligence.
MCP for local prospecting
The local-lead intelligence MCP server for your AI
Connect leadhoundapi (leadhoundapi.com) to your agent so it can read local markets, rank businesses with visible lead leakage, and draft outreach openers from public review signals, not guarantees.
Install
npm install -g leadhoundapi leadhoundapi auth login lh_live_... leadhoundapi mcp install claude --run
Hosted MCP clients can connect to https://api.leadhoundapi.com/mcp with OAuth or an account key.
Spend is capped
Agents use capped prepaid spend per call, job, and child key. No tool can create unlimited vendor cost.
Humans send
Keep a human on send. leadhoundapi gives evidence for outreach, not proof of intent.
Workflow transcript
the same agent that built it finds who needs it. Ask for a city, vertical, leak type, and opener. The agent reads only what the account can pay for.
User: Find appliance repair shops in Phoenix losing leads to missed calls. Agent: I will call leadhoundapi and cap spend to one paid market read. Tool: market scan returned public review-signal leakage and opener patterns. Agent: Top pattern is missed-call recovery plus warranty follow-up. Drafting 25 safe openers now. User: Export JSON and wait before sending. Agent: Export ready. I will not send outreach without approval.