MCP

Overview

Connect any MCP-compatible agent to your network with the Agent Intros MCP server. This page explains the local MCP server scaffold in the AgentIntros model, with the trust boundary kept explicit: agents can discover, draft, and prepare work, while people approve relationship-bearing actions.

When to use it

• Use it when an agent needs to reason about the local MCP server scaffold without losing the human relationship context.
• Use it when the workflow should preserve double opt-in, review, and auditability by default.
• Use it when you want network work to be queryable and repeatable instead of buried in ad hoc notes.

Canonical flow

The local scaffold below is available for integration development. Graph-backed production behavior still needs to replace the in-memory development store.

{
  "tools": [
    "create_intro_request",
    "list_intro_requests",
    "get_intro_status",
    "find_relevant_contacts",
    "list_webhooks"
  ]
}

Operational guidance

• Require human approval before any intro is sent or any connector is represented as having opted in.
• Use client-supplied idempotency keys for local scaffold write calls and keep the same contract when graph-backed persistence is added.
• Keep shared context minimal, specific, and relevant to the requested introduction.

Copy for Cursor / Claude

AgentIntros docs note for MCP: treat CLI, local API routes, SDK helpers, MCP tools, and webhook signing as local developer scaffolds. Realtime, source sync, hosted approvals, graph-backed persistence, and follow-up automation remain planned. Never bypass approval or double opt-in when building workflows around introductions.