Capabilities
You drive the Ema MCP with plain language. You describe what you want, and your assistant picks the right tool and calls it on your behalf, reading data, making a change, or chaining several steps together. This page groups what's available by area, with example prompts.
You don't need to know tool names. To see exactly what your assistant can do, ask "What Ema tools do I have access to?" or open the MCP view in your client (for example, /mcp in Claude Code or Codex). The exact set of tools depends on your role and Workspace, and grows over time.
What you can do
| Area | What you can do | Example prompt |
|---|---|---|
| AI Employees | List and inspect AI Employees (including their publish status: draft or published), create one from a template, from scratch, or by cloning, update configuration, set the EmaFusion™ model, and archive or restore. | "Create a Chat AI Employee called IT Support." |
| Workflows | Inspect an AI Employee's workflow, edit the graph (add or remove agents, edges, Tools, knowledge bases, and schedule agents), validate it, do a dry run, and publish a new version. | "Show me the workflow for the Sales Assistant AI Employee." |
| Knowledge bases | Create and list knowledge bases, upload and remove documents, search them, and manage data connections and sync. | "Upload this PDF to the Support knowledge base." |
| Integrations & Tools | Browse the integrations catalog, create and configure Tools, manage connections (including OAuth), and validate or invoke a Tool. Tools exposed via MCP carry integration metadata (such as product and docs URLs) so your assistant can describe them accurately; vendor migration details are never surfaced. | "Show me available integrations and add a Tool to the Sales Assistant AI Employee." |
| Governance & audit | Query the audit log to see who performed what action and when across your Workspace (including AI Employee changes, workflow publishes, and trigger state changes). | "Who published the HR AI Employee last week?" |
| Chat & testing | Start a chat session with an AI Employee, send messages, and review message history (useful for testing behavior). | "Send a test message to the HR AI Employee and show me the reply." |
| Dashboards | List dashboard tables and rows, upload a file, create and rerun rows, and read results. | "Upload this spreadsheet to the Invoice dashboard and run it." |
| Human in the loop | View pending human-in-the-loop requests, respond to them, and cancel them. | "Show me any approvals waiting on me." |
| Models | List the models available to your Workspace through EmaFusion™. | "Which models can I use?" |
| Debugging | Inspect recent runs and conversations to see what happened and why a step failed. | "What went wrong in the last conversation with the HR AI Employee?" |
| Identity | Confirm who you're signed in as and which Workspace you're working in. | "Who am I on Ema?" |
For background on the concepts these actions operate on, see AI Employees, Workflows, and Knowledge bases.
Reading versus changing
Some requests only read data: listing AI Employees, showing a workflow, inspecting a conversation. Others change your Workspace: creating, updating, publishing, or deleting. The MCP makes these changes as you, so they take effect immediately and are subject to the same permissions as the rest of the platform.
Review actions that change or delete. Creating, updating, publishing, archiving, and deleting are real changes to your Workspace. For example, archiving an AI Employee takes it out of service, and deleting a knowledge base, Tool, or connection is permanent. Most MCP clients ask you to confirm before running a tool that makes a change; read what the assistant proposes before you approve it. When you want to try something safely, connect to the Dev or Staging environment rather than Prod.
Workflow tools
The MCP gateway exposes tools for authoring workflow graphs directly. Two tools are relevant to scheduling and AI Employee listing:
workflow_dag_schedule_config_set: configures a schedule agent in a workflow DAG. Schedule agents are created bare (usingworkflow_dag_node_setwithtype: "schedule") and then configured with this tool, following the same bare-create-then-configure pattern ascall_aieandsend_emailagents.ai_employee_list: lists AI Employees in the Workspace. Each result now includes astatusfield indicating whether the AI Employee isdraftorpublished. Autopilot answers questions only from published AI Employees; drafts are listed so you can author them through the MCP, but their tools are not yet bound and they are not in service.
Scope and permissions
The connection always acts as the signed-in user, within one Workspace and environment:
- Your role decides what you can do. Building and publishing AI Employees requires builder access. If your role doesn't permit an action, the request is refused; the MCP can't elevate your access.
- Workspace isolation. You only see and change resources in the Workspace you signed in to. Resources in other Workspaces aren't visible.
- One environment at a time. A connection is pinned to the environment you set it up with. To act on a different environment, use that environment's connection.
For the full permission and role model, see Governance and Permissions.
What's next
- Setup — connect a client if you haven't yet.
- Troubleshooting and Limits — fix issues and understand the limits.
- Autopilot — the same capabilities, inside the Ema product.