DataHub and Data Gateway
DataHub connects source systems and prepares reusable datasets. A Data Gateway gives an approved application a bounded view of selected sources and released datasets. The gateway's grant determines what can be discovered and queried; team membership alone does not grant every connector or dataset.
Availability
Use a gateway provisioned and approved by your team. The source contracts described here do not establish that a gateway is deployed, that your account has access, or that every Uhuu product has adopted the same integration.
Set up the handoff
- The DataHub owner connects the source and confirms the objects, fields, and operations it supports.
- The owner grants a connected application the specific sources or released datasets it needs. An empty source grant means no sources. Dataset grants and any writeback grants are explicit and separate.
- The owner supplies the approved gateway connection. Its MCP address has the form
{datahub-origin}/v1/gateways/{slug}/mcp; use the actual supplied origin and slug. Keep credentials in the approved connection, never in a template, document payload, or browser example. - The application discovers the tools and reads their current input and output schemas before calling them. A dataset's own delivery endpoint and token are a separate access surface; do not substitute them for a connected application's gateway credential.
The gateway can expose live-source queries alongside tools for granted, released datasets. A dataset grant can pin a release revision or follow the production alias. A release captures its delivery schema; retain an explicit release identity and schema when a downstream flow requires reproducibility. A fixed schema does not freeze every source record, and live queries can change as records change.
Explore only what is granted
The live-source tool sequence is:
| Tool | Purpose |
|---|---|
list_sources | Discover the sources visible to this application. |
list_objects | Discover objects within a selected source. |
describe_object | Inspect the available fields and object description. |
get_source_capabilities | Check the supported query operations. |
query_records | Read a bounded page of records with supported filters. |
Follow the returned pagination and limits. A sample or one page does not establish that a dataset is complete, and unsupported filters must not be invented. Dataset tool names and schemas come from discovery rather than a hard-coded list.
Use data in a Uhuu document
The unified Uhuu MCP adapter exposes datahub__list_gateways, datahub__list_gateway_tools, and datahub__call_gateway_tool to a connection with datahub:read. It uses team-bound, configured gateways and exposes their tools marked read-only. It validates calls and declared result schemas; it does not expose gateway writeback or dataset authoring through this adapter.
For Documents MCP, select DataHub only when the template guide calls for it or the user explicitly requests an allowed source. Map returned facts into the template's agent-owned fields, retain relevant source or release provenance, and show gaps or assumptions before draft creation. Having a data scope does not make DataHub mandatory for every document.
For Template Studio, use the editor's supported data selection and save test data for a reproducible preview or handoff. A temporary browser preview is not a durable deployment record. Inspect the selected record and schema rather than assuming that every connected source is compatible with a template.
The result is a documented data selection with an explicit access boundary, which can support a reviewed draft or template preview. Writing back to a source or authoring a dataset requires its own supported, explicitly granted flow.