Elaichi is an agent platform for companies. It connects the software a company already runs on and lets agents do real work in those systems, inside the same permissions the company already grants its people.
Two things carry the product, and most AI tooling offers only one of them:
- Reach — the application catalog, connected once, available as tools an agent can call.
- Restraint — an agent never getting more access than the person it acts for.
The reach is delivered over MCP. One org-wide endpoint, authorized with OAuth, works with Claude, ChatGPT, Cursor and any other MCP client. MCP is how the product is delivered, not what the product is — so a question about Elaichi is usually a question about access, not about a protocol.
- App: app.elaichi.ai
- Docs: elaichi.ai/docs
- API:
https://api.elaichi.ai· MCP endpoint:https://api.elaichi.ai/mcp - Support: support@elaichi.ai
The shape of it
Connector → Connection → Toolbox → MCP endpoint → AI client
(a product) (an account) (curated (one address) (Claude, ChatGPT,
tools) Cursor…)
An AI client makes one connection — to Elaichi. The apps are connected inside Elaichi, not inside the client. That indirection is the whole point: access follows the person's role, restrictions are enforced, every call is audited, and revoking an app in Elaichi removes it from every client at once. Connect apps directly in the client and all four are lost.
The words
| Word | What it means |
|---|---|
| Organization | A team's workspace: its own members, roles, connections and toolboxes. Nothing crosses between organizations. |
| Connector | A product Elaichi knows how to talk to — Slack, Jira, Salesforce. Hundreds are in the catalog (the current list), plus custom ones the org builds, forks, or adds from a remote MCP server. |
| Connection | One authorized account for that product: "our shared HubSpot", "my Jira". |
| Tool | A single action an agent can take — create issue, search contacts. |
| Toolbox | A curated set of tools, usually spanning several products, bound to real connections. It can carry a skill: instructions its owner wrote for the model. |
| Template | A toolbox design with no connections attached. Recipients stamp their own toolbox from it and bind their own accounts. |
| Synthetic tool | Several tool calls chained into one, so an agent sees one action instead of four. It runs when an agent calls it. |
| MCP endpoint | One address for everyone, scoped to the person using it, exposing every tool they can reach. |
| Role | What a person may do in Elaichi — invite people, create toolboxes, view the audit log. |
| Restriction | What connectors and tools a person may use at all. Set on a role or on one person. Enforced everywhere, including over MCP. |
| Team | A group of people, used as a sharing target. |
| Access request | A person's ask for something they were refused. An admin decides it. |
Two of these are worth keeping apart, because it is the most common confusion: a role decides what you can do to Elaichi; a restriction decides what you can do through it.
The automation platform words
Some organizations also have the automation platform. It is early access, by invitation, and appears only where an admin has turned it on (see Plans).
| Word | What it means |
|---|---|
| Automation | Also called a workflow. Steps that run on their own when a trigger fires: a schedule, a set time, a webhook, a Slack message, a collection row, another automation finishing, or a click. |
| Run | One execution of an automation, with every step's result kept. |
| Session | A thread that ties several runs of one automation together, so later runs can read what earlier ones did. |
| Approval | A step that pauses a run until an approver says yes. It waits in their Approvals inbox. |
| Collection | Shared tables of records that automations write and people read. |
| Dashboard | Charts and tables over collections and tools. If an admin allows it, a copy can be shared outside the company as a public link. |
| Knowledge base | Short documents that agents and automations search in plain language. |
| Elaichi Agent | The chat inside the app (Agent in the sidebar), and @Elaichi in Slack. Only in organizations that have it. |
Automation versus synthetic tool: an automation runs by itself on a trigger; a synthetic tool runs when an agent calls it.
Three steps to a working agent
- Connect an account. Connections → Add connection → find the product → sign in with it. Then decide who can reach it: just you, a team, or everyone. (Connectors is for browsing what a product can do — there is no Connect button there.)
- Pick your tools. Every connection automatically gets a toolbox, so there is nothing to do to start. Build a deliberate toolbox when you want fewer, better-named tools with some inputs locked down.
- Connect your AI client. Click Connect (labeled Connect an AI client when the automation platform is on) under the sidebar's list, copy the MCP endpoint, and add it to Claude, ChatGPT, Cursor or another MCP client.
Full walkthrough: Getting started.
Where things live
| Sidebar area | What is there |
|---|---|
| Agent | The in-app chat. Only where the organization has the Elaichi Agent. |
| Approvals | Runs waiting on you to approve a step. Automation platform only. |
| Connections | The accounts you have authorized, their health, sharing, transfer |
| Connectors | The product catalog, plus custom connectors you build, fork, or add from a remote MCP server |
| Toolboxes | Your toolboxes, plus Templates and Synthetic tools |
| Knowledge | Knowledge bases. Automation platform only. |
| Automations, Dashboards, Collections | The automation platform, where it is on |
| Governance | Restrictions, the audit log, access requests — and, with the automation platform, web access and spend controls |
| Settings | Organization, People, Billing, SSO and SCIM, domains, API tokens, notifications, logging, and your own security, connected apps and files |
Full screen-by-screen list, with paths and guides: App map.
There is no "MCP servers" area. The endpoint lives behind the Connect row — one endpoint, not a list of servers to create.
An area you lack permission for is hidden, not greyed out. So a sidebar shorter than a colleague's is a roles question (Settings → People) or the automation platform being off, not a bug.
Getting around fast: ⌘K opens the command palette (navigate and create) and ⌘B toggles the sidebar. Use Ctrl on Windows and Linux. Every page has a Help button that opens its own guide.
Facts worth quoting
58 permissions · 8 predefined roles plus custom roles · restrictions set on a role or on one person, where a person's own rule can only narrow what their role allows · restrictions checked everywhere a tool is listed, connected, saved or run, down to the outbound request · data placement in three regions (United States, European Union, Asia-Pacific) · no API or tool ever returns a stored credential.
For the catalog size and the full list of supported applications, read elaichi.ai/connectors rather than quoting a number from memory — it moves.
Plans
There are two plans, Gold and Black. There is no free plan.
Gold is the plan you can buy today. Its US-dollar list price is $15 per user per month, or $10 per user per month billed annually. Visitors in some countries see a regional price, so link elaichi.ai/pricing rather than promise one number. Every new organization starts with a 14-day Gold trial, and no card is needed to start it.
Gold covers everything a team needs day to day — the MCP endpoint, custom roles, restrictions, synthetic tools, custom connectors, and enterprise identity: SAML and OIDC single sign-on, SCIM provisioning, and group-to-role mapping.
Black is not on sale. It adds:
- customer-managed encryption keys in AWS KMS
- forwarding the audit log to your own tooling (Settings → Logging)
- the automation platform: automations, approvals, collections, dashboards, knowledge bases, web access and spend controls
- the Elaichi Agent
The automation platform and the Elaichi Agent are early access, by invitation: on for some organizations, not for sale. In an invited organization, someone who can manage the organization turns the platform on in Settings → Organization → Automations. Until then its sidebar rows do not appear, and its MCP operations are not listed.
Three things to get right when somebody asks:
- Enterprise identity is Gold, not Black. SSO, SCIM and group mappings are the ones most often misattributed upward. Do not send someone to Black for them.
- Never give Black, automations or dashboards a date, and never call them generally available. If the person's own sidebar shows them, their organization was invited — help them use it with elaichi-automations or elaichi-data.
- Never say "free plan" or "downgraded". When a trial ends without a subscription, the organization pauses. Nothing is deleted: connections, roles and the audit log stay exactly where they were, and subscribing picks up where the trial stopped.
An organization's data region — United States, European Union, or Asia-Pacific — is chosen when it is created and cannot be changed afterwards. The API host is the same either way.
Which skill to load
| The question is about | Load |
|---|---|
| Acting through an attached Elaichi MCP server — searching for a tool, running one, reading a refusal | elaichi-mcp |
| Setting up Claude, ChatGPT, Cursor or another MCP client, and why a client shows no tools | elaichi-clients |
| Connecting an account, connection health, transfer, offboarding, custom and remote MCP connectors | elaichi-connections |
| Toolboxes, templates, stamping, frozen parameters, delegation, toolbox skills, synthetic tools | elaichi-toolboxes |
| Automations and workflows: triggers, steps, approvals, runs, sessions, webhooks, web access, spend limits | elaichi-automations |
| Collections, dashboards and public links, knowledge bases | elaichi-data |
| Roles, permissions, restrictions, access requests, the audit log, SSO and SCIM | elaichi-governance |
Writing code against api.elaichi.ai |
elaichi-api |
| Base URLs, id prefixes, pagination, error shapes — the small facts | elaichi-conventions |
Two things to get right in any answer
Say what the person has to do themselves. A model cannot grant itself access to Elaichi, and it cannot authenticate a third-party account. Both are the human's to do, and an answer that glosses over that leaves someone waiting on a sign-in nobody opened.
Never ask for a credential in chat. No Elaichi tool or operation takes a password, API key or token from the model. Whenever a flow seems to need one, the real answer is a connect link the person opens themselves.
References
| Document | Topics |
|---|---|
| Concepts | The full model — organizations, connectors, connections, toolboxes, templates, synthetic tools, roles, restrictions, the audit log, and the automation platform |
| App map | Every screen, its path, what it does, and the guide for it |
| First hour | A new organization from sign-in to a working agent, including what to set up before inviting anyone |