Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.xpertai.cn/llms.txt

Use this file to discover all available pages before exploring further.

Agent Tools are the stable execution interfaces that the UOSE system provides to agents. They wrap resources, entities, relationships, actions, and governance capabilities into a fixed set of tools so Agents can complete tasks within a controlled semantic space.

​
Tool List

Core tools include:
  • queryEntities: query candidate entities by intent and scope.
  • getEntityNeighborhood: get the one-hop neighborhood of the target entity and attributes of related entities.
  • queryOntologySchema: view entity types, relation types, and action types of a resource.
  • queryRdf: execute RDF debugging queries when allowed.
  • discoverActions: discover actions executable on the target object.
  • simulateAction: simulate and validate an action before execution.
  • executeAction: execute an action that passed validation.
  • getAuditTrace: view the audit trace of a task.
These tools are exposed through REST APIs and can also be exposed to business Assistants by an MCP provider. Agents should use this main path: 
queryEntities -> getEntityNeighborhood/queryOntologySchema -> discoverActions -> simulateAction -> executeAction -> getAuditTrace
This path has these benefits:
  • Locate the object first, then read context.
  • Discover actions first, then construct parameters.
  • Simulate and validate first, then execute for real.
  • Review audit evidence after execution.
Agents should not skip discoverActions and execute directly, and should not choose by themselves when the entity is not unique.

​
Target Selection

Action tools use target to select objects. Prefer reusing the preferredTarget returned by the previous step, which contains:
  • entityRef
  • entityTypeCode
Use entityId only when copying the full UOSE UUID. Do not treat ontology ID fragments, short hashes of external keys, or display names as entityId.

​
MCP Access

The MCP provider exposes UOSE Agent Tools as constrained tools and derives these values from trusted headers:
  • tenant.
  • organization.
  • actor identity.
  • assistant identity.
  • current business user token.
For resources that use current_user_oidc, MCP calls must forward the business user token, for example through x-uose-actor-token. Otherwise, although the Agent has an MCP server token, it cannot access BI or xpert-pro source systems on behalf of the user.

​
Tool Output

Tool outputs are kept as Agent-consumable as possible:
  • discoverActions returns allowed actions, denied actions, reasonCode, and input schema.
  • executeAction returns status, executionId, policyDecision, auditRef, and effect.
  • MCP output omits some large fields, such as summary, appliedPlan, and audit, and indicates them through omitted fields.
Agents should plan the next step from structured fields instead of parsing natural language logs.