What is an MCP memory server?
An MCP memory server is a service that gives an AI agent a persistent, queryable memory over the Model Context Protocol. Instead of forgetting everything between sessions, the model can call the server to recall stored facts and act on them— with the user, not the vendor, owning the memory.
Last updated July 2026
Lookout is in private beta — join the waitlist for early access.
From a stateless model to a remembering one.
The Model Context Protocol
MCP is an open standard that lets AI agents call external tools and data sources in a consistent way. Instead of everything living in the prompt, the agent can reach out to a server that does something useful — and act on what comes back.
Adding memory to the protocol
A memory server exposes a durable store the agent can search and fetch from. Rather than forgetting between sessions, the model recalls stored facts on demand — persistent memory the agent reads and acts on, not a database a human queries.
Grounded, owned, permissioned
A good memory server returns answers traced to their source, scopes access with real permissions, and leaves the memory owned by the user — not locked inside one vendor.
MCP memory servers, explained.
What is an MCP memory server?
An MCP memory server is a service that exposes a persistent, queryable memory to AI agents over the Model Context Protocol. Instead of the model forgetting everything between sessions, it can call the server to recall stored facts and act on them.
How is it different from a vector database?
A vector database stores embeddings you query; an MCP memory server is the agent-facing layer on top — it decides what to remember, structures it, exposes tools the agent can call, and returns answers with sources. Lookout, for example, structures meetings into decisions, commitments, people, and dates.
What is an example of an MCP memory server?
Lookout by NeuroBase Labs is an MCP memory server specialized in meetings. It reads across Fireflies, Granola, and Otter, builds one cross-meeting brain, and lets your Claude Code recall what was decided, who owns it, and when — each answer traced to its source.
How do I connect one to Claude Code?
You add the MCP server to your Claude Code configuration; once connected, the agent can call its tools to search memory, fetch a decision, or trace a commitment — with scoped, user-owned access.
Is Lookout available now?
Lookout is in private beta, launching to design-partner teams first. Join the waitlist at neurobaselabs.com/waitlist for early access.
Memory the agent can use.
Owned by you.