Skip to content

Product tour

The whole path, from a connected app to a call you can read back

Connect an app, compose a toolbox, point any MCP client at one endpoint, and govern every call. Four moves, each one drawn below.

Start 14-day free trial Talk to us

Connect

Connect an app once, and its tools exist

One hosted sign-in puts the credential in a vault nobody reads back, and every documented method of that app becomes a tool an agent can call. There is nothing to build.

  • Day one. 450+ connectors, owned by a person, a team, or the whole company.

  • Anything else. Describe an internal API in JSON and it joins the catalog beside the rest.

  • Forks. Fork a public connector, then pull upstream changes when you want them, not when they land.

One application picked out of the catalog, one sign-in, and that application’s own tools on the far side.

Compose

A toolbox is what you hand to a team

Pick the tools a team should have, rename them so the model reads them the way you meant, and share the capability without sharing the credential.

  • Pinned values. Fix an argument to one value the model can neither see nor change.

  • Several apps, one call. Chain the steps server side, through the same checks and the same log.

  • Nothing to set up. Every person gets a toolbox per connection the moment that connection goes live.

Five connected tools, one switched off, and three toolbox entries out. The last entry is a single tool that runs across two applications.

Authorize

One address for the whole company

Point Claude, ChatGPT, Cursor, any MCP client, or the Elaichi Agent at one endpoint and sign in with OAuth. No token to paste, no second URL to keep track of.

https://api.elaichi.ai/mcp The same for every user.

  • Live. Read from the source, at the moment of the call.

  • Revoke in one click. Disconnect a client and its session ends. Nothing to rotate, nothing left behind.

  • Big sets stay quiet. Past about thirty tools the endpoint switches to search, so a large catalog never floods the model.

One person and one endpoint, with every client signing in through it rather than holding an address of its own.

Govern

Three checks before a call leaves, and a record after

Roles decide who may act, restrictions decide which tools exist for them at all, and every call that does run lands in an append-only log.

  • Never advertised. A restricted tool is not refused when it is called. The model is never shown it.

  • Person over role. A rule set on one person replaces the one their role carries.

  • Your log too. Forward the audit trail to your Datadog, batched and retried.

  • Offboarding. Removing a member makes you transfer or delete what other people were leaning on. Never a silent break.

Three gates, left to right: the person’s role, the restrictions on them, and any pinned argument. The middle call stops at the restriction and never reaches the application. The two that pass land in the log.

Elaichi Agent

An agent that works inside the same rules as everyone else

It calls the same tools as any other client, with the access of the person it acts for, and anything that writes waits for that person to say yes.

  • Your models, your bill. Anthropic, OpenRouter, Fireworks, or any OpenAI-compatible gateway.

  • Approval as a step. Allow once, always allow, or deny, on the action itself.

  • It runs Elaichi too. Create a toolbox, share it, connect an account, through the same checks a person gets.

The agent is another client of the same endpoint, drawn inside the same boundary and holding the access of the person it acts for.

Roadmap

From answering questions to doing the work

Automations and dashboards are launching soon. When they land, the rules do not change: the same permissions, and the same log.

Automations

The work happens without anyone asking

A trigger starts a sequence of steps: call a tool, reshape the result, branch, wait, and ask a person when it matters. Only one step is a model, so the rest stays fast.

Elaichi digest

Run 418 · started 2 minutes ago · on behalf of Emma Laurent

  1. ✓

    Schedule

    Every weekday at 08:00

    0.2s
  2. ✓

    Fetch records

    Elaichi

    1.4s
  3. ✓

    Group by owner

    Transform

    0.1s
  4. ✓

    Draft the digest

    Agent step

    Ran with 4 tools, returned a structured summary

    6.2s

  5. Approve the digest

    Needs approval

    Assigned to Michael Brennan

    Approve

  6. Post the digest

    Elaichi

    Queued

Collections and dashboards

The morning report builds itself

Collections hold data pulled from your apps and whatever an automation writes down. Dashboards read from them and refresh on a schedule rather than on every view.

Elaichi health

Refreshed 4 minutes ago · every 15 minutes · from the records collection

Live

Records

1,284 ↓ 12%

Reports

96 ↓ 8%

Needs attention

3 ↑ 2

Updated this week

412 ↑ 9%

Records created

Last 14 days

By team

Share of activity

Engineering 34%

Product 27%

Support 21%

Sales 18%

Put agents to work on your own systems

14 days on Gold, no credit card. Start with one app and one team.

Start free trial Talk to us
Works with
Claude ChatGPT Cursor and any other MCP client, or the Elaichi Agent.
When the trial ends
Nothing is deleted. Connections, roles and the audit log stay where they are, so subscribing picks up exactly where you left off.