Internal alpha · CLI/package live · graph verbs in scaffold

Relationship context for your agent.

AgentIntros turns your professional network into a first-class, composable tool surface for Claude Code, Codex, Cursor, and your own workflows: warm paths, source provenance, strength signals, and draftable context without letting an agent act on your behalf.

Install the CLI Read the docs

published intros CLI

MCP scaffold

/api/v1 scaffold

browser approval gate

intros-mcp / relationship graphscaffold

"Who at a16z do I already know through someone I have worked with?"

toolagentintros.find_path

targeta16z fintech investor

strength0.81

sourcesGmail · Calendar

agent searches

find_path

graph returns

2-hop path

context

get_context

action boundary

approval

Current surfaces

Built for agent-native operators who want their network available where the work already happens.

CLI

intros

Initialize, connect sources, query the graph, and print MCP config.

npm i -g intros

MCP

intros-mcp

Expose relationship-context tools inside Claude Code, Codex, Cursor, and custom agents.

find_path

API

/api/v1

Use canonical graph verbs in custom workflows.

get_context

Approval

human gate

Automate read and draft; gate sends, exports, sharing, and key rotation.

pending_review

The problem

Your network is offline from your agent.

Agent-native founders already trust their AI tools with code, docs, and workflows. The relationship layer is still trapped in unstructured places, so the agent cannot reliably answer who can help, why the path is warm, and what should happen before an ask.

Relationship data is scattered.

Warm paths are buried across Gmail, contacts, calendar context, enrichment sources, fallback files, and memory.

Agents cannot reason about warm paths.

Claude Code and Codex can inspect code and call APIs, but they cannot infer who you know, how warm the path is, or why the request is appropriate.

Warm-path work is still manual.

You search LinkedIn, ask a founders group, scan old email, and draft from scratch. The work is high leverage, so the operational drag compounds.

The solution

Your agent finds
the path. You
control the action.

Add one entry to your MCP config. Now your agent can query a relationship-context graph: warm paths, source provenance, strength signals, enrichment disclosures, and draft scaffolds that stop before relationship-bearing action.

Step 1 / 10

~/agent · claude-code · mcp:agentintros

userWho at a16z do I already know through someone I have actually worked with? Give me context and a draft scaffold, but do not send anything.
▍

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

Developer proof

One path from CLI to approval.

The technical promise is compact: seed trusted data, expose canonical graph tools to the agent, and stop before relationship-bearing action.

Install

npm install -g intros

Connect or import

Google-first sources; CSV/JSON fallback

Expose MCP

Paste the generated block into your agent

Approve action

Relationship-bearing steps stay browser-gated

agentintros-contract.ts

// Canonical v1 tool surface
tools = [search, find_path, get_context, list_strongest, suggest_nurture, draft_intro]

// Approval boundary
agent can read_context + draft_scaffold
browser approves send | export | share | delete | rotate_key

Sources

GmailContactsCalendarCSV / JSON

Runtimes

CLIMCPSDKAPI

Tools

searchfind_pathget_contextlist_strongestsuggest_nurturedraft_intro

Boundary

Agents read context and prepare drafts. Browser approval gates send, export, share, delete, and key rotation flows.

Internal alpha posture: no borrowed logos and no fake applause. Evaluate the live scaffold, bounded contract, and approval behavior.

Comparison

Different tools for
different relationship jobs.

This is not a takedown. Happenstance, Dex, and CRMs solve real problems. AgentIntros is narrower: a personal-first relationship graph for agents that can find paths, draft context, and stop at human approval.

AgentIntros fits when

Your agent is already part of your work, and your network should be a private queryable graph with a strict approval gate.

Happenstance fits when

Your main need is a team-shared connection pool for sales, recruiting, or fundraising discovery.

Dex fits when

You want a polished personal CRM for remembering people, notes, reminders, and manual follow-up.

A CRM fits when

You need account ownership, pipeline reporting, forecasting, and manager-visible sales process.

Buyer criterion

AgentIntros

Happenstance

Dex

CRM

Agent surfaceCan an agent call it directly?

strong

AgentIntros: CLI, MCP, SDK, and REST are core surfaces.

strong

Happenstance: Has MCP and targets agent-native workflows.

not focus

Dex: Mostly human-facing personal CRM.

partial

CRM: APIs exist, but workflows are pipeline-first.

Graph postureWhat relationship model is emphasized?

strong

AgentIntros: Personal graph with provenance, strength, recency, and agent-readable context.

partial

Happenstance: Team graph and connection discovery.

partial

Dex: Contact timeline and reminders.

partial

CRM: Accounts, contacts, deals, and activities.

Send boundaryWho takes responsibility for the message?

strong

AgentIntros: Browser approval gates sends, sharing, export, deletion, and key rotation.

partial

Happenstance: Evaluate send and approval semantics per account.

strong

Dex: Human-authored outreach is normal.

partial

CRM: Often supports automation and sequencing.

Agent surface

Can an agent call it directly?

AgentIntros

strong

AgentIntros: CLI, MCP, SDK, and REST are core surfaces.

Happenstance

strong

Happenstance: Has MCP and targets agent-native workflows.

Dex

not focus

Dex: Mostly human-facing personal CRM.

CRM

partial

CRM: APIs exist, but workflows are pipeline-first.

Graph posture

What relationship model is emphasized?

AgentIntros

strong

AgentIntros: Personal graph with provenance, strength, recency, and agent-readable context.

Happenstance

partial

Happenstance: Team graph and connection discovery.

Dex

partial

Dex: Contact timeline and reminders.

CRM

partial

CRM: Accounts, contacts, deals, and activities.

Send boundary

Who takes responsibility for the message?

AgentIntros

strong

AgentIntros: Browser approval gates sends, sharing, export, deletion, and key rotation.

Happenstance

partial

Happenstance: Evaluate send and approval semantics per account.

Dex

strong

Dex: Human-authored outreach is normal.

CRM

partial

CRM: Often supports automation and sequencing.

Alpha note: AgentIntros rows describe the launch direction plus current scaffold where noted.

Install path

Install the CLI.
Connect sources.
Paste MCP.

The current developer path is intentionally plain: install the published `intros` package, create local auth, connect Google-first sources as they land, or import CSV/JSON fallback data, then paste the MCP block into Claude Code, Cursor, Codex, or any MCP-capable agent.

Install

npm install -g intros

Initialize

intros init creates local auth config

Connect or import

Google-first; CSV/JSON remains fallback

Connect your agent

intros mcp-config prints the block to paste

CLI reference MCP setup SDK reference

intros@0.1.0· CLI + MCP scaffold live

Claude Code · Cursor · Codex

scaffold

configGenerated after init. Paste it into the agent you already use.

{  "mcpServers": {    "agentintros": {      "command": "intros-mcp",      "env": {        "AGENTINTROS_API_KEY": "ait_dev_..."      }    }  }}

sources

Gmailm1Google Contactsm1Google Calendarm1BrightData LinkedIn enrichmentm1X enrichmentm1CSV / JSON fallback

Graph routing, MCP access, and approval-gated review are the launch contract. Today, the safe alpha loop is graph setup, path query, context retrieval, draft scaffold, and browser approval before any relationship-bearing action.

What works today?

The published CLI/package, MCP binary, API scaffold, SDK contracts, browser auth, and dashboard shell are live.

Is this another CRM?

No. AgentIntros is a relationship-context graph your agent can query, not a contact database you manually maintain.

Can an agent send messages?

No relationship-bearing send, export, sharing, deletion, or key rotation happens without browser approval.

Why alpha now?

The first proof is behavioral: useful paths, source provenance, draft scaffolds, and a hard approval boundary.

Start here

Put your network where your agent can ask.

Install the CLI, connect or import relationship context, then paste the MCP config into the agent you already use.

Start alpha setup npm i -g intros

Relationship context for your agent.

intros

intros-mcp

/api/v1

human gate

Your network is offline from your agent.

Relationship data is scattered.

Agents cannot reason about warm paths.

Warm-path work is still manual.

Your agent findsthe path. Youcontrol the action.

One path from CLI to approval.

Install

Connect or import

Expose MCP

Approve action

Different tools fordifferent relationship jobs.

AgentIntros fits when

Happenstance fits when

Dex fits when

A CRM fits when

Install the CLI.Connect sources.Paste MCP.

What works today?

Is this another CRM?

Can an agent send messages?

Why alpha now?

Put your network where your agent can ask.

Your agent finds
the path. You
control the action.

Different tools for
different relationship jobs.

Install the CLI.
Connect sources.
Paste MCP.