MCP agent wrapper tools¶

MCP exposes complete specialist subgraphs, not the individual tools registered inside Oria. Both wrappers accept user_application_id, user_message and an optional debug=false.

opportunities_suggestion_agent_mcp_tool¶

The wrapper authenticates the MCP HTTP request, resolves tenant database and AI configuration, loads the requested user and verifies that the opportunity agent is enabled. It handles the shared /clear_history command before graph execution, then invokes the persisted user/tenant thread with only search_opportunities_tool allowed.

The subgraph interprets natural language, extracts structured criteria and runs the opportunity search described in the Oria tool catalogue. For Smart Mobility, the wrapper resolves the request language through the shared py3langid then tiny LLM chain before invoking the subgraph. If both are inconclusive, it keeps its business-locale fallback. Language detection does not access the MongoDB translation cache. The response contains final text, structured opportunity cards, agent name and success status. Typed card data is cleared from the graph state after it is returned so it is not replayed on the next call. Unexpected execution errors produce a stable unsuccessful response.

talent_finder_agent_mcp_tool¶

The wrapper performs the same authentication, user loading, tenant-agent availability and persistent-thread setup for the Talent Finder subgraph. Its specialist uses the same request-language resolution, converts the request into structured intent and searches users while excluding the caller. The MCP result is the final agent text; execution failure returns Error processing your request..