Clients setup
Connect an external MCP client to a supported DataMaker deployment.
The current DataMaker server uses Streamable HTTP, with separate metadata and execution endpoints. The in-product agent already connects to these services. An external client needs the endpoint URL and a credential for tool calls; there is no supported one-command install for all clients.
Before configuring a client
Obtain these from your DataMaker administrator:
- The metadata endpoint, normally the API host plus
/mcp. - Any execution or system endpoint you need, such as
/mcp-exec. - A credential for tool calls: an API key with the role the client needs, or a supported way to obtain and refresh a session JWT for that deployment. An API key does not expire with a session, so it suits a client that runs unattended.
- The authorized team and project IDs.
Send one credential with the request context:
X-API-Key: <api-key>
X-Team-Id: <team-id>
X-Project-Id: <project-id>or, with a session:
Authorization: Bearer <session-jwt>
X-Team-Id: <team-id>
X-Project-Id: <project-id>If a client sends both, the API key is used. These are HTTP headers, not a complete client configuration file. Store keys and session tokens using your client's secret-input mechanism. Do not copy browser cookies into a shared config or commit credentials.
Register and verify
- Add a remote MCP server using your client's Streamable HTTP transport and the supplied URL.
- Configure the authentication and scope headers using the client's supported settings. If it cannot send the required headers or refresh sessions, use DataMaker chat or REST/SDK automation instead.
- Connect and inspect the announced tool list. Listing needs no credentials, so seeing the catalogue does not prove the credential works. Register additional execution or system services separately when required.
- Ask it to list templates in the selected project. This is the first call that uses the credential. An empty list can be valid for a new project.
- Test a small generation only after confirming the project and available tool schema.
A successful metadata connection does not prove an execution service is configured. Desktop execution may require a local service and access to the target from that machine.
Troubleshooting
| Symptom | Check |
|---|---|
Tool call fails with status: 401 | No valid credential reached the call. Check the API key or session token, its expiry and the API host. The error appears inside the tool result; the MCP request itself still succeeds, and listing tools keeps working without credentials. |
Tool call fails with status: 403 | Team membership, project scope, the API key's role or the user's permissions, and pending approvals. A READ_ONLY key cannot call tools that write or run. |
| Generation tool absent | Inspect both the connected metadata and execution catalogues for your release. |
| Old install command fails | The legacy standalone npm package is unpublished; use the current services described above. |
| Tool schema changed | Reconnect to refresh the catalogue and use the advertised arguments. |
For automation that does not need an MCP client, use REST, Python or TypeScript.