Compliance & Audit
Audit trails, data handling policies, and compliance posture.
Audit trail
MarcoPolo records data operations performed through the gateway. The Activity page shows when an operation ran, the MarcoPolo workspace identity and connection involved, the action, and its outcome. Depending on the operation, details may include the MCP tool name, command or query text, row count, and duration. Users can review their own activity; administrators can review activity across their organization and export the events shown.
Activity and tracing. Activity events are stored in the OTEL Logging service. Request payloads are captured in the OTEL Tracing service and displayed in Activity details. Tool response payloads are not captured in traces by default because they may contain customer data.
Workspace artifacts provide another record of work done when they are saved. You can browse query files under connections/<name>/queries/, files pulled into data/downloads/, and saved scripts, analyses, and generated outputs in your workspace.
Provenance
Data provenance lets you trace an AI answer to the information behind it: the data sources accessed, the queries or commands run, and, where available, the specific records or document sections used. For interactions routed through MarcoPolo, Activity identifies the workspace identity, time, tool or operation, data connection, and outcome. The associated trace provides the request context needed to examine what was submitted.
When an organization needs to establish exactly what data a tool returned, MarcoPolo can support capturing tool response payloads in the OTEL Tracing service as part of its provenance record. Before enabling response capture, the organization should decide which responses to retain, whether sensitive fields need redaction, who may inspect the traces, and how long the data should be kept.
Data handling
No data replication. MarcoPolo operates as a passthrough layer. Queries execute directly against your connected systems. The results you pull are cached in your workspace (DuckDB) for iteration, but your source data is never replicated, warehoused, or centralized by MarcoPolo.
No training. Your data, queries, and workspace contents are never used for AI model training. This applies to all data that flows through MarcoPolo.
Workspace persistence. Your workspace files persist across sessions: cached query results, downloaded files, generated scripts, and artifacts. Source data stays in your systems; what persists is the results and files you chose to pull into your workspace.
Data residency
MarcoPolo workspaces run on cloud infrastructure. For enterprise customers with specific data residency requirements, contact [email protected] to discuss deployment options: regional workspace deployment, VPC peering and PrivateLink, or Bring Your Own Cloud (BYOC).
Common questions
Can the AI or MarcoPolo see my data? The AI only sees query results you explicitly request. It has no background access to your systems.
Is my data used for AI training? No. Customer data never enters any training pipeline.
Can MarcoPolo employees access my workspace? Access to customer workspaces is restricted to authorized support operations with audit logging.
What happens if I delete a connection? The connection is removed. Cached results from previous queries may still exist in your workspace's DuckDB, but no new queries can be made against the removed connection.
Download the security brief (PDF)
For compliance documentation or security questions, contact [email protected].