AgentIntros
npm i -g introsStart free
Live with Claude Code, Codex & Cursor

Your network
in the palm of
your agent.

AgentIntros gives your AI agents access to the people in your world. It's the intelligence layer for building relationships and surfacing warm intros.

Start for free npm i -g intros
cliagentintros.find_path
awaiting your approval
query
"Find a warm path to a seed investor who's already met someone in my network."
AgentIntros relationship graph showing a user's AI agent using the AgentIntros tool to find warm paths to operators, investors, connectors, and designers.
DROPS INTO YOUR STACK ↓
Claude Code·Codex·Cursor·Hermes·OpenClaw
§ 05 / Install

Two minutes.
One new tool
in the stack.

No new interface. No new habit. Add the MCP server to your agent config and your network becomes a tool call.

CLI reference APIDocs
STATUSapi.agentintros.ai · operational · p50 38ms
Wire into Claude Code, Cursor, Codex
{  "mcpServers": {    "agentintros": {      "command": "intros-mcp",      "env": {        "AGENTINTROS_API_URL": "https://api.agentintros.ai",        "AGENTINTROS_API_KEY": "ai_..."      }    }  }}
Sources you connect once
GmailCalendarLinkedInSOONXSOONNotionSOONContacts
§ 01 / What's underneath

Three layers,
one address.

It's not AI for your contacts. It's your contacts, structured for your AI: a graph, a vector store, and a relational record, all queryable behind a single MCP server.

I
Graph layer

Who knows who, and how strongly.

Your agent traverses the warmest path between any two people, weighted by relationship strength, recency, and mutual edges. The graph is the API.

0 ms median lookup
II
Vector layer

Context, searchable by meaning.

Every email, meeting, and touchpoint is stored as embeddings so your agent finds the right person by what they actually care about, not just who they know.

Hybrid BM25 + ada
III
Relational layer

Your full history, always structured.

The record that backs every graph path and semantic search: queryable rows your agent can cite, never a black box.

Postgres + agentic-db
§ 02 / Watch your agent work

Your agent calls.
You approve the send.

AgentIntros exposes a handful of clean tools over MCP. Your agent picks the right one, queries your graph, and routes the result back for your approval. trust is the product.

Step 1 / 8
~/agent · claude-code · mcp:agentintros
userI need to raise a seed round. Find the warmest path to a fintech-friendly investor, and tell me where I can give value first.
▍

ⓘ  Real conversation skeleton. Tools are auto-discovered from the MCP server.

§ 03 / Tools, not features

Every relationship job,
as an MCP tool call.

You shouldn't browse a graph. Your agent should. AgentIntros exposes a small surface so any LLM can pick the right call.

sales

"Find warm paths to design partners at mid-market fintech companies."

→ mcp toolagentintros.find_warm_path
recruiting

"Find a founding ML engineer who has shipped frontier-model infrastructure."

→ mcp toolagentintros.find_relevant_contacts
fundraising

"Find seed investors who already know us through a trusted operator."

→ mcp toolagentintros.find_warm_path
give first

"Who in my network could I add real value to right now, before I need to ask?"

→ mcp toolagentintros.find_giving_opportunity
nurturing

"Who haven't I talked to in 90 days that I'd want to keep close?"

→ mcp toolagentintros.list_stale_relationships
bd

"Companies where we have warm relationships AND a real reason to partner."

→ mcp toolagentintros.find_warm_path
§ 04 / The honest landscape

Not a nudge engine.
Not a contact list.
Not a referral ledger.

The category that exists today was built for humans to browse. AgentIntros is built for agents to query. That's the whole shift.

Capability
AgentIntros
Happenstance
Dex / GoodWord
Graph-native routing
YES
NO
NO
Native MCP for agents
YES
NO
NO
Give-first prompts
YES
NO
NO
No manual upkeep
YES
YES
YES
Approval-gated sends
YES
YES
YES

Graph-native routing

AgentIntrosYES
HappenstanceNO
Dex / GoodWordNO

Native MCP for agents

AgentIntrosYES
HappenstanceNO
Dex / GoodWordNO

Give-first prompts

AgentIntrosYES
HappenstanceNO
Dex / GoodWordNO

No manual upkeep

AgentIntrosYES
HappenstanceYES
Dex / GoodWordYES

Approval-gated sends

AgentIntrosYES
HappenstanceYES
Dex / GoodWordYES

ⓘ Based on public product surface as of 2026. Corrections welcome.

§ 06 / Common questions

Honest answers.
No marketing detours.

No. AgentIntros indexes your sources (Gmail, Calendar, LinkedIn, etc.) automatically and exposes the result as a graph + vector + relational layer for your agent. There is no UI for you to maintain.

Happenstance nudges you to reach out. Dex is a personal CRM you tend by hand. AgentIntros routes through your graph, finds the warmest path, surfaces a give-first opportunity, and hands the draft back to your agent and to you for approval.

Your agent finds paths and drafts context. You approve the send. That's a feature, not a limitation. Trust is the product, and approval-gating is how it stays intact.

Postgres through an AgentIntros domain fork of agentic-db: pgpm deployment, PostGraphile/GraphQL access, vector search, full-text search, and graph-ready relationship tables in one database.

Your network is yours. Data is stored per-user, encrypted at rest, and never used to train a model. You can wipe everything via a single CLI call.

Founders and operators already running Claude Code, Codex, or Cursor daily. If your agent is the first surface you open every morning, your network is the obvious next thing to wire in.

Coming for everyone. Built for agent-native founders today.

Put your network where your agent can reach it.

Start for free. Bring your own MCP-capable agent.

Start for free Install CLI