Reasoning doesn't execute.
APIs and MCP expose system state, decision logic, and execution routes.
Every action is defined. Every step is logged.
Every action has a schema. Every response is validated.
Agents operate through defined interfaces. Inputs and outputs are enforced.
Profile → Risk → Portfolio → Compliance → Advisor → Client. Each hop is a defined interface call. Timed. Logged. Auditable.
No black box. No free-text chains. Inputs in. Outputs out. At every step.
Regulatory regimes require reasoning to be evidenced, not narrated. The MCP transcript is that evidence.
Returns portfolio state — positions, risk, constraints — at a fixed timestamp.
Scores a signal against portfolio context. Returns probability, urgency, confidence, expected impact.
Tests an action against mandate, exclusions, suitability tier, and regulatory rules.
Routes the action to the venue — broker, custodian, advisor queue, or compliance hold.
Reconstructs a historical decision from frozen inputs and call transcript. For audit.
Captures suitability inputs. Produces a validated profile. Hands off to downstream agents.