Skip to content

MCP tools

AI Hub currently registers three streamable-HTTP MCP tools:

Tool Purpose
health_check verify MCP availability
opportunities_suggestion_agent_mcp_tool run the opportunity specialist subgraph
talent_finder_agent_mcp_tool run the talent-finder specialist subgraph

The two agent wrappers accept user_application_id, user_message and an optional debug flag. They verify that the associated agent is enabled for the tenant before invoking it.

See agent wrappers for their inputs and workflow.

Authentication

MCP HTTP requests require:

X-API-Key: <api-key>

AI_HUB_USER_API_KEY and AI_HUB_ADMIN_API_KEY are accepted credentials. If both are absent or empty, MongoDB fallback is available only without DATABASE_URL; it prefers the API key's clientId and falls back to Client-ID, while using the fixed smartmobility data source. PostgreSQL uses the single database configured by DATABASE_URL; after resolving the user, the agent receives that user's setting.id only for business-data filters.

State

Agent-wrapper threads use the tenant and user IDs, share the application's LangGraph checkpointer, and support /clear_history.

The former navigation entries for direct user, opportunity, objective, feedback, expert and skill MCP modules were removed because those modules are not registered in the current repository.