MCP Memory Extension: Persistent Agent Memory

What Is the MCP Memory Extension?

The MCP Memory Extension provides persistent memory storage and retrieval capabilities for AI agents — enabling agents to remember context across sessions, learn user preferences over time, and maintain awareness of in-progress work. Without it, every agent session starts with a blank slate.

Four Types of Agent Memory

• Session memory — context maintained within a single conversation. This is provided by the LLM's context window and doesn't require an extension.
• User preference memory — preferences and patterns specific to an individual user, stored across sessions. The agent learns how a user prefers to work.
• Task state memory — the state of in-progress work. When a multi-step task spans sessions, the agent needs to know where it left off.
• Knowledge memory — durable facts the agent should always know: company structure, key contacts, frequently referenced data.

Memory Governance

Persistent agent memory raises governance questions that stateless agents don't: what is stored, for how long, who can access it, and how is it deleted when the user requests it or the retention period expires?

Define memory retention policies before deploying agents with persistent memory. At minimum: retention period per memory type, access control (who can see what an agent has remembered about a user), deletion process (GDPR and similar requirements), and audit logging for memory read/write operations.

Implementation in Workato

Workato's MCP Memory Extension exposes memory as a set of MCP tools: store_memory, retrieve_memory, list_memories, and delete_memory. The agent calls these tools to manage its own memory — the extension handles the persistence layer, encryption, and access control.