This server provides durable, user-owned personal memory for Claude Code on Walrus, with login plus save, search, analyze, and restore capabilities.
memwal_login: Opens a browser wallet flow and stores credentials locally so other memory tools become usable.
memwal_remember: Saves a fact or memory the user explicitly asks to remember, storing the full detailed text without summarizing.
memwal_recall: Searches the user's Walrus Memory for facts relevant to a query, returning matches ranked by relevance.
memwal_analyze: Extracts memorable facts from a passage (preferences, habits, biographical info, constraints) and saves each as a separate memory.
memwal_restore: Re-indexes a namespace from Walrus blobs back into the relayer's search index, with pagination and truncation handling.
Provides tools for interacting with Walrus Memory, a decentralized memory service on the Sui blockchain, allowing agents to remember, recall, analyze, and restore memory entries with Sui wallet-based authentication.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Walrus Memory MCP Remember that I prefer dark mode"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Walrus Memory gives Claude Code durable, user-owned memory. This plugin packages the published @mysten-incubation/memwal-mcp server with setup guidance, slash commands, and lifecycle hooks.
The MCP runtime, tests, and npm releases stay canonical in MystenLabs/MemWal/packages/mcp. This repository is plugin packaging only.
Publisher: Mysten Labs. Dashboard: memory.walrus.xyz. Docs: Claude Code setup.
After the plugin is listed in the Claude plugin directory, install Walrus Memory from /plugin → Discover.
To load this repo locally for review:
claude plugin validate . --strict
claude --plugin-dir .Then restart Claude Code or run /reload-plugins, and confirm /plugin and /mcp show memwal.
MCP-only (tools, no plugin hooks):
claude mcp add --scope user memwal -- npx -y @mysten-incubation/memwal-mcp@0.0.13 --label "Claude Code"Related MCP server: LedgerMem MCP Server
/memwal:setupmemwal_login opens the browser wallet flow and stores credentials at ~/.memwal/credentials.json (mode 0600). Do not paste that file or its private key into chat.
Connect Walrus Memory and confirm memwal_health is ok.
Remember that I use pnpm and TypeScript strict mode in this project.
What package manager and TypeScript settings do I prefer?
/memwal:setup
/memwal:health
/memwal:remember
/memwal:recall
/memwal:analyze
/memwal:restore
/memwal:logout
This plugin uses local stdio MCP and delegate-key login. It does not use the hosted Claude custom-connector OAuth flow.
The hosted connector is a separate remote MCP surface at https://relayer.memory.walrus.xyz/api/mcp. See docs/usage/hosted-connector.md.
Walrus Memory encrypts memories with SEAL and stores them on Walrus. The local MCP process keeps a delegate private key on disk after login so it can sign relayer calls.
Privacy policy: https://docs.wal.app/docs/legal/privacy
Product docs: https://docs.wal.app/walrus-memory/mcp/claude-code.html
Support: open an issue on this repository, or use the Walrus Memory dashboard
Apache-2.0
Extract memorable facts from a passage of text (preferences, habits, biographical info, constraints) and save each as a separate Walrus Memory memory.
Inputs: text (required), namespace.
Sign this MCP client into your Walrus Memory account by opening a browser. Run once when the agent reports Walrus Memory is not signed in. Opens the dashboard in the default browser, waits for wallet approval, then writes credentials to ~/.memwal/credentials.json. Other memwal_* tools become usable on the next call after a successful login.
Inputs: no named parameters.
Search the user's Walrus Memory for facts relevant to a query. Returns matching memories ranked by relevance.
Inputs: query (required), limit, namespace.
Save a fact to the user's Walrus Memory personal memory. Call ONLY when the user explicitly asks to remember/save something. Pass the full, detailed text — never summarize.
Inputs: text (required), namespace.
Re-index a namespace from Walrus blobs back into the relayer's search index. Returns counts and truncated status; call again with a higher limit when truncated=true.
Inputs: namespace (required), limit.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 5 tools
Each tool serves a distinct purpose: login for authentication, remember for explicit saving, analyze for batch extraction, recall for searching, and restore for index maintenance. Although analyze and remember both create memories, their trigger conditions are clearly separated (passage processing vs. direct user request), preventing misselection.
All tools follow a uniform memwal_<verb> pattern using lowercase snake_case. The verb stems (analyze, restore, login, remember, recall) are consistent and predictable, making the naming highly coherent.
With only five tools, the surface is compact and focused on the essential operations of a personal memory system: authentication, writing, reading, batch processing, and maintenance. No tool feels redundant or extraneous.
The set covers authentication, creation, recall, and bulk ingestion, but lacks update and delete operations. For a memory system, the inability to forget or correct stored facts is a notable gap that could limit agent effectiveness.
Governed personal world model and memory for your AI agent. Pair once, connect over MCP.
Hosted agent memory: store, search, and recall facts across sessions from any MCP client.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Cross-session memory for AI agents with a Source Receipt for every memory, over MCP.
