Skip to content
← back to projects
2026·shipped·Forward Deployed Engineer

Agentic Event Store

An event-sourced ledger on PostgreSQL with CQRS projections, optimistic concurrency, cryptographic integrity, and MCP exposure so LLMs query history without bypassing the audit log.

10 Academy TRP · 1-week brief
PythonPostgreSQLCQRSEvent SourcingMCP
Code ↗

Pipeline

1
Event log
Hash-chained, append-only
↓
2
CQRS projections
Rebuilt from the log
↓
3
Integrity checks
Tamper-evident
↓
4
MCP tools
Agent-queryable

Highlights

  • ▸Cryptographic hash-chained events: you can prove the ledger has not been tampered with.
  • ▸Event upcasting handles schema evolution without rewriting historical events.
  • ▸Exposed to agents through MCP tools so LLMs can query history without side channels.

Why event sourcing matters here

Agentic systems produce actions that need to be auditable after the fact. If an agent granted access or moved money, you need the full causal chain, not just the current state. Event sourcing gives you that chain by construction.

Design highlights

  • Optimistic concurrency on the aggregate root: writes fail fast when the stream has moved.
  • CQRS projections rebuilt from the log, so read models can be dropped and rebuilt without data loss.
  • Cryptographic integrity via a hash chain, where each event commits the hash of the previous one.
  • Event upcasting so old events can be interpreted under new schemas.

MCP integration

The store exposes an MCP tool set: read_stream, query_projections, verify_integrity. Agents can inspect their own history without touching the raw database, and every read is itself an event.