MCP (Model Context Protocol) — Overview
AI agents query AI CDP as a tool via MCP endpoints. All calls require consent verification and produce audit logs.
Architecture
AI Agent → MCP Server → Consent Check → Service Layer → Response + Audit
Security
- Consent checking before any profile access
- PII masking in all responses
- Rate limiting per agent (configurable)
- Full audit logging of every tool call
Available Tools
| Tool | Parameters | Returns | Consent |
|---|---|---|---|
get_profile | dtx_id, fields[] | Profile subset | MCP_ACCESS |
search_profiles | criteria, limit | DTX_IDs (no PII) | ANALYTICS |
get_segment_members | segment_id, sample_size | Sample DTX_IDs | ANALYTICS |
resolve_identity | identifier_type, value | DTX_ID | MCP_ACCESS |
get_segment_stats | segment_id | Aggregate stats | None |
create_segment | name, criteria | Segment ID | Admin only |
get_identity_graph | dtx_id, depth | Related entities | MCP_ACCESS |
See MCP Tools Reference for detailed schemas.