MCP server
Understand DataMaker’s built-in MCP services and external-client requirements.
DataMaker exposes tools through the Model Context Protocol (MCP). The in-product agent is configured with the services it needs automatically; you do not need to install a separate MCP package to use chat.
Current architecture
The DataMaker API hosts Streamable HTTP MCP endpoints. Metadata and execution are separate tool surfaces:
| Endpoint | Purpose |
|---|---|
/mcp | Discover and manage assets, and generate from a saved template (generate_from_id). |
/mcp-exec | Execution and data operations: scenarios, data jobs, masking, key maps, endpoint reads and writes. It carries no template generation tool. |
/mcp-plan | Create and patch reviewable plans. |
/mcp-exec/plan-run | Report the progress of a plan run already under way: logs, files, deliveries and completion. It does not start a run. |
/mcp-exec/{namespace} | System tools, such as SAP, Tosca, Atlassian and Signavio, when registered. |
The configured execution host can differ from the metadata API host in desktop team mode. Use the URLs supplied by your deployment, not an assumed public hostname for every service.
Authentication
Listing tools does not require credentials. A tools/list request to any endpoint above returns that endpoint's catalogue, so a client can discover the tools before access is configured.
Calling a tool does require credentials, because each tool runs the matching DataMaker API request as the caller. Authenticate tool calls with either:
- a session JWT in
Authorization: Bearer, or - a REST API key in
X-API-Key. If both are sent, the API key is used.
Send X-Project-Id and X-Team-Id for the project and team the tools act on. The catalogue is the same for every caller; what a call may do depends on the credential. A session acts with the user's team permissions. An API key acts within its own scope and role: a READ_ONLY key is refused by tools that need write or run permissions, a READ_WRITE key can also write and run but not administer the team or reverse masking, and an ADMIN key can do both. See API keys.
For external clients, your administrator provides the endpoint and a credential. See Clients setup.
Older standalone package
The separate datamaker-mcp repository describes an older stdio server. Its package @automators/datamaker-mcp is currently unpublished on npm. The old pipx install datamaker-mcp and unscoped npx datamaker-mcp instructions are not a supported setup for the current built-in server.
Use DataMaker chat for the integrated experience, or the REST API and SDKs for automation that does not need an MCP client.
Working with tools
Start with read-only discovery: list the active project's templates or inspect a saved plan. Read the announced tool schema before invoking it; tool names, arguments and output shapes differ across services. Review the target and data volume before writes, and honor any approval request.
See Tool catalogue for the main tool groups.