MCP Servers
The MCP servers tab of the Integrations page lets you connect external Model Context Protocol servers so the tools they expose become available to your AI Employees. You point Ema at an MCP-compatible server; Ema discovers the tools it advertises, an admin activates the server, and from then on those tools can be used like any other Tool.
The sequence is short: add the server → Ema discovers its tools → an admin activates it → add a connection (credentials) → its tools are callable.
Add an MCP server
From the MCP servers tab, select Add MCP server and fill in:
- Name — a display name for the server.
- Description — what the server's tools do.
- Endpoint URL — the MCP server's URL (for example
https://api.example.com/v1/mcp). - Authentication — how Ema authenticates to the server (see below). Defaults to None.
Select Add server. Ema connects to the endpoint, discovers the tools it advertises, and sets the server to Pending review — no tools are usable until an admin activates it. Adding the server registers it and discovers its tools; you supply any credentials afterward, when you add a connection.
Ema connects to MCP servers over Streamable HTTP. The endpoint must be a Streamable-HTTP MCP server — SSE and stdio transports aren't supported.
Authentication types
Pick how Ema authenticates to the server:
| Type | What it is |
|---|---|
| None | The server needs no credentials. |
| API key | Ema sends an API key. |
| Bearer token | Ema sends a bearer token. |
| Basic auth | Ema sends a username and password. |
| OAuth 2.1 (Authorization Code) | Users sign in through the server's provider; Ema runs the OAuth flow and refreshes tokens automatically. |
Add a connection
Adding the server is metadata only — it holds no credentials. To supply them, add a connection under the server and choose a connection type:
- Shared — one connection the whole workspace uses.
- User — each user connects their own account.
Then enter the credentials for the server's authentication type — an API key, a bearer token, or a username and password. For OAuth 2.1, you don't paste a token: Ema discovers the server's authorization server, registers itself as a client automatically where the server supports it, sends each user through the provider's consent screen, and refreshes access tokens on its own.
Activate and manage the server
An MCP server moves through a deliberate lifecycle:
- Pending review → an admin selects Activate to move it to Active.
- Active → its tools are callable. An admin can Disable at any time.
- Disabled / Discovery failed → an admin can Enable, which re-runs discovery and returns the server to Pending review.
From the actions menu you can View tools (the discovered tool list), Refresh tools (re-discover from the server), and Delete the server — deletion is blocked while an active AI Employee still uses it. Editing the server's URL or credentials re-runs discovery and returns it to Pending review, so a changed endpoint never silently serves stale tools.
Use a server in another workspace or environment
There's no export/import step for MCP servers. To use the same server in another workspace or another environment, just add it again there — with the same endpoint URL and authentication. Registration is lightweight, and its tools are re-discovered when you add it.
Ema can also expose its own AI Employee tools to MCP-aware clients through a separate MCP gateway — the reverse direction (Ema as an MCP server). That's documented with the MCP product, not here.
What's next
- Data Connectors — sync external documents into Knowledge bases so an AI Employee can cite them.