Ema MCP
The Ema MCP server lets you manage your AI Employees from outside the Ema platform. Connect an external AI assistant — Claude Desktop, Claude Code, Codex, or any other MCP client — to your Ema tenant, then describe what you want in plain language and the assistant does it for you: listing and inspecting AI Employees (AIEs), building and editing workflows, managing knowledge bases, connecting Tools, and debugging conversations.
It exposes the same family of platform capabilities that power Autopilot inside the product — but to the AI tool you already work in, rather than the in-app panel.
What is MCP? The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external systems. An MCP server exposes a set of tools; an MCP client (your AI assistant) calls them on your behalf. The Ema MCP server is hosted by Ema — there is nothing to install or run locally.
What you can ask for
Once connected, you talk to your assistant in plain English. For example:
- "List all my AI Employees."
- "Create a Chat AI Employee called IT Support."
- "Show me the workflow for the Sales Assistant AI Employee."
- "What went wrong in the last conversation with the HR AI Employee?"
If you're not sure where to start, ask "What can I do on Ema?" or "What Ema tools do I have access to?" and your assistant will list what's available. See Capabilities for the full picture.
Supported clients
The Ema MCP server uses streamable HTTP transport with OAuth, so it works with any compliant MCP client. The clients we document setup for are:
- Claude Desktop app — usable in Chat, Cowork, and Code modes.
- Claude Code — the terminal coding assistant.
- Codex — the OpenAI command-line assistant.
Other MCP clients that support streamable HTTP and OAuth work the same way; point them at the environment URL and authenticate. See Setup.
Environments
The server is available for four Ema environments, each a separate deployment with its own sign-in and data:
| Environment | MCP endpoint |
|---|---|
| Dev | https://dev.ema.ai/api/v1/agent-harness/mcp |
| Staging | https://staging.ema.ai/api/v1/agent-harness/mcp |
| Prod | https://app.ema.ai/api/v1/agent-harness/mcp |
| Demo | https://demo.ema.ai/api/v1/agent-harness/mcp |
Connect to the environment where your AI Employees live. To work across environments, add a separate connection for each. Estimated setup time is about a minute.
Access. You sign in with your own Ema account, so the connection acts as you: it can only see and change what your role and tenant allow. You need an account with builder access to create or edit AI Employees. If you can't sign in or don't have access, contact your administrator.
In this section
| Page | What it covers |
|---|---|
| Setup | Connect Claude Desktop, Claude Code, Codex, or another client, authenticate, and verify the connection. |
| Capabilities | What you can do through the MCP, by area, with example prompts and a note on safe, scoped actions. |
| Troubleshooting and Limits | Fix authentication, missing-tools, and wrong-environment issues, and understand the current limits. |
What's next
- Setup — connect your client in about a minute.
- Capabilities — everything the MCP can do for you.
- Autopilot — the same kind of help, built into the Ema product.