# Connect MCP connectors once, use every LLM | Elaichi

> Connect each SaaS account once and use it from Claude, ChatGPT, Cursor, and any MCP client through one OAuth endpoint. No per-client reconnection.

Elaichi gives your organization one MCP endpoint, at the same address for every organization: https://api.elaichi.ai/mcp. Connect each SaaS account once, then point Claude, ChatGPT, Cursor or any MCP client at that address and sign in with OAuth. Each person connects once, with no token, secret or client ID to paste, and when you adopt a new client your accounts are already there, because they are connected to Elaichi rather than to one chatbot.

- **1** address for Claude, ChatGPT, Cursor and every other MCP client
- **0** MCP servers to build, host or keep updated
- **0** tokens, secrets or client IDs to paste
- **1** search tool to find every connected app’s tools

## How connecting usually goes

- Getting an app into AI at all usually means finding, hosting, or building an MCP server for it, and deciding where its keys live.
- Most tools connect to one AI client at a time, so Salesforce, Gmail, Smartlead, and Slack each get set up inside ChatGPT.
- Adopt another client later, say Claude, and the same accounts are set up again in a different place.
- Across a whole team and every client people want to try, the same connections get rebuilt in many places, with tokens kept wherever each client expects them.

## One endpoint, every client

- Connect each account once, to Elaichi. Any client you add later signs in to the same endpoint over OAuth, with no separate setup per client.
- Point Claude, ChatGPT, Cursor or any other MCP client at one address. Clients register themselves, so beyond that one address there is no token, secret or client ID to paste.
- Apps outside the catalog, including your own internal APIs, come in as custom connectors built from a JSON config, served from the same endpoint as everything else.
- Connected tools are never listed one by one. The model finds them with search_tools and runs them with execute_tool, so its tool list stays the same size however many apps you connect.

## Built for teams that use more than one AI

- **People who switch between clients:** You work in ChatGPT, then Claude, then Cursor, and want the same apps ready in each one without setting them up again.
- **IT and platform teams:** You want to roll out AI clients across the company without a server, token or setup per person per client to maintain.
- **Engineers and builders:** You need internal APIs and apps outside the catalog reachable from the same place as everything else, under the same rules.

## What it looks like in practice

### The same accounts in ChatGPT and Claude

*Sales.* An account executive connects Salesforce, Gmail and Slack once. They prospect in ChatGPT in the morning and review deals in Claude in the afternoon, and every account is already there in both, because neither client holds the connection.

Connectors: [Salesforce](https://elaichi.ai/connectors/salesforce/), [Gmail](https://elaichi.ai/connectors/gmail/), [Slack](https://elaichi.ai/connectors/slack/)

### Cursor without a server per developer

*Engineering.* An engineering team adds the one Elaichi address to Cursor. Each developer signs in with their own grant and reaches Jira and Sentry through it, with no MCP server to run per person and nothing to update when a connector changes.

Connectors: [Jira](https://elaichi.ai/connectors/jira/), [Sentry](https://elaichi.ai/connectors/sentry/)

### An internal API beside the SaaS apps

*Platform.* A platform team adds their internal billing API as a custom connector from a JSON config. It answers through the same endpoint as every catalog app, under the same roles, restrictions and audit log.

## How to set it up

1. **Connect your apps.** Pick apps from the catalog and sign in through the hosted flow, or add a custom connector for anything else. Each account is connected once, to Elaichi.
2. **Add the address to each client.** Use https://api.elaichi.ai/mcp everywhere. On Claude Team and Enterprise an owner adds it as a custom connector for the organization, on ChatGPT Business an admin adds it as an app, and in Cursor it goes in the MCP settings.
3. **Each person signs in once.** Members connect and sign in with their own OAuth grant, choosing on Elaichi’s consent screen what the client may read, change, run or delete.
4. **Work across clients.** The same connected accounts answer in every client. Disconnect one client under Settings, Connected apps, and the others keep working.

## Three ways to get your apps into every AI

| | Each client’s own connectors | A server per person | Elaichi |
|---|---|---|---|
| Setup | Each app, again in each client | A server per app, often per person | Each app once, to Elaichi |
| Adding an AI client | Connect everything again | Point it at each server and configure it | Add one address and sign in |
| Credentials | Held separately by each client | Kept in each server’s configuration | In the vault; clients hold an OAuth grant |
| Apps outside a catalog | Only what each client offers | Write and run your own server | A custom connector from a JSON config |
| Who maintains it | Each vendor, on its own schedule | Your team | Elaichi authors and runs its catalog connectors |
| Switching a client off | Disconnect each app in that client | Revoke keys in each server | Disconnect it; its next request is refused |

## Details that matter

- **One address for everyone:** https://api.elaichi.ai/mcp is the same for every organization. What changes from person to person is the grant, clamped by roles and restrictions, not the URL.
- **Nothing to paste:** Clients register themselves through OAuth dynamic client registration with PKCE, so there is no client ID, secret or header to enter.
- **Connectors, already built:** Elaichi authors and runs the connectors in its catalog, so the apps your team already uses are ready to connect, and a team can also bring its own remote MCP server under the same rules.
- **Scopes you choose:** On the consent screen each person picks whether a client may read, change, run tools or delete, and delete is never pre-ticked. A client can be narrowed to chosen toolboxes, up to 50.

## Further reading

- [Connect company apps to Claude and ChatGPT](https://elaichi.ai/blog/connect-company-apps-to-claude-and-chatgpt/)
- [Connect Elaichi to Claude](https://elaichi.ai/blog/connect-elaichi-to-claude/)
- [Connect Elaichi to ChatGPT](https://elaichi.ai/blog/connect-elaichi-to-chatgpt/)
- [Connect Elaichi to Cursor for a whole team](https://elaichi.ai/blog/cursor-mcp-one-endpoint-vs-per-developer/)
- [The MCP context window problem, and a fix](https://elaichi.ai/blog/context-window-problem-mcp-tools/)
- [Running your own MCP servers: the real cost](https://elaichi.ai/blog/self-hosted-mcp-servers-vs-control-plane/)

## Frequently asked questions

### Do I have to reconnect my accounts for each AI client?

No. You connect each account once to Elaichi. Claude, ChatGPT, Cursor and any other MCP client point at the same endpoint and sign in over OAuth. Switching clients does not reconnect anything.

### What address do I add to my client?

https://api.elaichi.ai/mcp, the same for every organization. Copy it exactly, with no trailing slash and no query string. In the web app, Connect your AI client shows it with a copy button.

### Which AI clients work with Elaichi?

Claude, ChatGPT, Cursor and any other MCP client. Each person connects once and signs in with their own OAuth grant.

### Can I connect an internal tool or an app that is not in the catalog?

Yes. Custom connectors are authored from a JSON config and served from the same endpoint as every other app. You can also fork a public connector and pull its later changes through a review step.

### Is there a separate URL or token per toolbox?

No. There are no per-toolbox URLs and no embedded tokens. The endpoint is a single address for the whole organization; what varies per person is their grant, not the URL.

### Does connecting many apps flood the model with tools?

No. Connected tools are never listed one by one, however few there are. The model finds them with search_tools and runs them with execute_tool, so the list it sees stays the same size however many apps you connect.

### What happens when I disconnect a client?

Disconnecting it under Settings, Connected apps ends that client’s grant, and its next request is refused. Your other clients and your connected accounts are untouched.

### Can I use it from CI or a headless agent?

Signing in to Elaichi happens in a browser, so it works from clients a person signs in to, such as Claude, ChatGPT and Cursor. A client running headless in CI, with no browser, cannot complete the OAuth sign-in.

## Related use cases

- [Context sharing across teams](https://elaichi.ai/use-cases/context-sharing/)
- [Governance](https://elaichi.ai/use-cases/governance/)
- [Automated provisioning and deprovisioning](https://elaichi.ai/use-cases/provisioning/)

See also [Product](https://elaichi.ai/product/) and [Security](https://elaichi.ai/security/), or [start a trial](https://app.elaichi.ai/signup).