AI Agent Signal Sources currently submit trading intent through vorda.propose_action. Scoped account visibility, history, previews, and controlled actions are the planned MCP operations surface.
Sandbox lets you watch an agent work with paper fills and preview flows under real validation rules.
Key takeaways
- MCP is the transport for an AI Agent Signal Source, not a separate execution path.
- Current availability starts with vorda.propose_action.
- Planned tools remain bounded by workspace permissions, Automation readiness, exact Trading Rules, reconciliation, and Execution Proof.
What works now
Create an AI Agent Signal Source, connect a supported MCP client, and submit structured trading intent through vorda.propose_action.
A safe source test validates and displays the normalized proposal. It cannot place a trade.
The planned operations surface
Vorda plans scoped tools for balances and positions, execution history and failures, order previews, and policy-controlled order or position actions.
Clients must discover tools at runtime. A planned tool must never be described as available before it is released and authorized.
The boundaries do not move
The AI client never receives broker credentials, chooses an unauthorized broker, changes Trading Rules, or bypasses a stopped, incompatible, or unready Automation.
Every executable action remains attributable to its Signal Source, immutable Rules version, target account, Automation, reconciliation result, and Execution Proof.
FAQ
Answers users search for before connecting automation.
Can an agent inspect positions today?
Not through the current AI Agent Signal Source. Scoped account visibility is planned and will appear only when the corresponding MCP tools are released.
Can an agent execute directly?
No. Current agents propose. Future controlled action tools still require explicit authorization and a compatible, ready, running Automation.