MCP Record Action Tools: AI Clients Can Now Discover and Execute Record Actions

The MCP Standard Tools SuiteApp now includes two record action tools that expose NetSuite's existing record-action framework to AI clients over the Model Context Protocol. These tools do not create new actions, nor do they support record macros or bulk actions — they surface and execute the actions that already exist on a given record.

New tools

• ns_getRecordActionMetadata — Returns the list of actions available for a record type or a specific record instance. Results are filtered by the calling user's role, the record type, and the record's current state.
• ns_executeRecordAction — Executes one of those actions against a specific record. The caller must supply the target record, the action identifier, and any required parameters.

Key constraints

• Role-scoped: Only actions the authenticated role can see are returned. If permissions change between the metadata call and the execute call, the action may become unavailable.
• State-dependent: Available actions depend on the record's current state. A record that has transitioned (e.g., a sales order moving from Pending Approval to Approved) may expose a different action set than when metadata was last retrieved.
• No idempotency guarantee: Oracle warns explicitly against automatic retries. Executing the same action twice can apply the change twice (e.g., double-approving, double-posting). AI agent implementations must treat these calls as non-idempotent and require confirmation before execution.
• No bulk support: Each execution targets a single record. There is no batch endpoint.

Example: Bank Reconciliation Suggestion

Oracle references Bank Reconciliation Suggestion record actions as a concrete example. This implies actions like accepting or rejecting reconciliation suggestions are available through these tools, letting an AI assistant walk an accountant through bank-rec workflows programmatically.

What the source does not say

The documentation is thin on implementation detail. The following are not specified and should be verified in the SuiteApp's own documentation or by calling ns_getRecordActionMetadata against test records:

• The full list of record types that expose actions through this interface.
• The parameter schema for ns_executeRecordAction (whether parameters vary per action, what validation is performed).
• Governance cost per call.
• Whether these tools respect SuiteScript beforeSubmit / afterSubmit user-event triggers, or workflow action scripts, when the underlying action mutates the record.
• Error response structure when an action is no longer available.

What to do

1. If you are building or maintaining an AI agent integration with NetSuite via MCP: Add ns_getRecordActionMetadata and ns_executeRecordAction to your tool registry. Always call the metadata endpoint first to discover available actions before attempting execution.
2. Implement confirmation gates. Do not allow your AI agent to auto-execute record actions without human confirmation. The non-idempotent nature of these calls means a retry loop could double-post transactions or duplicate approvals.
3. Handle stale-state errors. Between the metadata call and the execute call, the record may have changed. Build your agent to gracefully re-fetch metadata if an execute call fails due to an unavailable action.
4. Audit your roles. The actions surfaced depend on the role associated with the MCP integration's credentials. Review which actions each integration role can see, especially for sensitive record types like journal entries, vendor bills, or employee records.
5. Test against your custom record actions. If you have created custom record actions via SuiteScript or workflows, verify whether they appear in the ns_getRecordActionMetadata response. The documentation does not clarify whether custom actions are included.