Skip to content

Getting started

Connect the software your company already uses to the AI clients your team works in — Claude, Cursor, Dust, and anything else that speaks MCP.

You authorize an account once, choose which tools to expose, and give each person their own endpoint. Nobody shares credentials, and administrators keep control of what agents can reach.

The words we use

Term What it means
Connector A product Elaichi knows how to talk to, such as Slack, Jira, or Salesforce
Connection One authorized account for that product — “our shared HubSpot”, “my Jira”
Tool A single action an agent can take, like create issue or search contacts
Toolbox A curated set of tools, often spanning several products
MCP server Your personal endpoint that exposes a toolbox to an AI client

MCP (Model Context Protocol) is the open standard AI clients use to discover and run tools. Elaichi speaks it, so anything that supports MCP works without custom integration code.

How the pieces connect 
flowchart LR
  Connector["Connector<br/>product catalog"] --> Connection["Connection<br/>authorized account"]
  Connection --> Toolbox["Toolbox<br/>curated tools"]
  Toolbox --> MCP["MCP server<br/>your endpoint"]
  MCP --> Client["AI client<br/>Claude, Cursor, Dust…"]

Sign in

Open app.elaichi.ai. There are no passwords — sign in with:

  • Google, GitHub, or Microsoft
  • A magic link sent to your inbox
  • Your company’s single sign-on, if an administrator has set it up

Your account is global. The first time you sign in you either create an organization, accept an invite, or join automatically because someone verified your email domain.

An organization is your team’s workspace: its own members, roles, connections, and toolboxes. Nothing crosses between organizations.

Three steps to your first agent

First-agent path 
flowchart TD
  A["1. Connect an account"] --> B["2. Pick tools<br/>connection toolbox or shared template"]
  B --> C["3. Create an MCP server<br/>from My toolboxes"]
  C --> D["Paste URL into your AI client"]

1. Connect an account

Open Connectors, find the product you need, and choose Connect. Sign in with that product and approve the access it asks for.

Decide who owns the connection: just you, a team, or the whole organization. Credentials go into a secure vault and are only used when a tool runs.

2. Pick your tools

Every connection automatically gets a toolbox, so you can start immediately.

When you want something more deliberate, build a shared toolbox: pick the handful of tools that matter for a job, rename them so an agent picks the right one, lock down inputs you don’t want it choosing, and share it with your team. Everyone who uses it brings their own connections.

3. Create an MCP server

From one of your own toolboxes, create an MCP server. You get an address and a token — copy both, because the token is shown once — and paste them into your AI client’s configuration.

That client now sees your tools and can run them as you, with your permissions.

Keeping it safe

  • Roles decide what each person can do in the app.
  • Restrictions decide which products and tools a role may use at all, and apply everywhere — including anything reached through MCP.
  • The audit log records meaningful changes, and can forward events to your own logging tools.