Walrus Memory brings cross-tool AI memory via a single MCP prompt
Cole Medin demos Walrus Memory, an MCP-based memory layer that stores encrypted, user-owned AI context outside any single tool, letting agents like Claude Code and Pi share the same memories instantly.
Score breakdown
Walrus Memory removes the lock-in of tool-specific memory systems, allowing context created in one AI coding assistant to be recalled immediately by a completely different agent without any re-setup.
- 01Walrus Memory is an MCP server-based memory layer that stores AI context outside any single tool or provider.
- 02Setup requires only a single prompt copied from the Walrus homepage, pasted into any AI coding assistant.
- 03Memories are encrypted and tied to a key owned by the user, not stored on a provider's infrastructure.
Cole Medin describes a workflow problem he encountered while switching between multiple AI coding assistants — Claude Code, Codex, and Pi — for different tasks: memory built into one tool doesn't carry over to another. His solution is Walrus Memory, a cross-tool memory layer that operates as an MCP server and stores memories outside any single provider's infrastructure.
Walrus is set up with a single prompt copied from its homepage, which configures the MCP server inside any AI coding assistant.
Walrus is set up with a single prompt copied from its homepage, which configures the MCP server inside any AI coding assistant. Once connected, the MCP server exposes tools for saving individual memories, analyzing larger text blobs to extract key facts, and searching stored memories. In the demo, Medin pastes a large block of text into Claude Code and instructs it to use Walrus to remember the contents; the agent calls the MCP server and extracts 20 facts, which are encrypted and stored. A second Claude Code session then queries those memories successfully, and the same query run inside Pi — with only the MCP server connected and no other setup — returns the same information.
All memories in Walrus are encrypted and tied to a key the user owns. The system supports access control through delegate keys, allowing different agents or people to access memories in different ways, with the ability to revoke access at any time. Walrus also offers an SDK for building the memory layer into custom applications. The video was produced in partnership with Walrus.
Key facts
- 01Walrus Memory is an MCP server-based memory layer that stores AI context outside any single tool or provider.
- 02Setup requires only a single prompt copied from the Walrus homepage, pasted into any AI coding assistant.
- 03Memories are encrypted and tied to a key owned by the user, not stored on a provider's infrastructure.
- 04In the demo, Claude Code extracted 20 facts from a pasted text blob and stored them via the Walrus MCP server.
- 05A Pi session retrieved the same memories with no additional setup beyond connecting the MCP server.
- 06Users can create delegate keys to control which agents or people can access specific memories, and revoke access at any time.
- 07Walrus also provides an SDK for integrating the memory layer into custom applications.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 20, 2026 · 08:55 UTC. How this works →