
Memorix MCP Server: Setup for Claude, Cursor and Other Apps
Gives Claude, Cursor, Codex and other coding assistants one shared project memory that survives new chats and tool switches.
Memorix lets your coding assistant save decisions, fixes and gotchas, then recall them in later chats and in other assistants. It stores notes in a local SQLite database tied to your Git project, and the server needs no account or API key. The default server shows nine tools. Larger sets (up to 48) switch on by setting MEMORIX_MODE. The package also installs a terminal agent and a setup command that adds hooks, so read what setup changes before you run it.
Good for
- Developers who re-explain the same project to every new chat
- People who switch between Claude Code, Cursor and Codex on one repo
- Teams that want decisions and fixes kept next to the code
- Privacy-minded users who want notes kept on their own machine
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install Memorix MCP
Claude Code
- 1
Copy this prompt.
Add the Memorix MCP server to Claude Code. Run this command: claude mcp add memorix -- npx -y memorix serve Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Node.js 22.18 or newer, Git.
Read-only mode: There is no read-only switch. Tell the assistant not to store memories, or run the micro tool set and review what it stores.
The generated commands add serve because plain memorix opens its terminal agent instead of the server (README). The server starts with its smallest set of 9 tools. Memorix stores notes under the Git project you open, so start your assistant from inside a Git repo. The setup command is different: it writes memorix serve --mode lite into the app's MCP config, so that route shows 21 tools, not 9. It also registers a local plugin (memorix@memorix-local) whose hooks run memorix hook on SessionStart, UserPromptSubmit, PostToolUse, PreCompact and Stop, and adds guidance to ~/.claude/CLAUDE.md. Add --noHooks to skip hook capture and run memorix uninstall --dry-run to see what would be removed.
Built from the package data we hold. Not yet tested by us.
Tools
Builds a short brief for the task you are about to start
Assembles a compact brief with start-here files, memories tied to the code, warnings about stale notes and ways to verify. It can refresh the local code index first.
Try: “Get the project brief before we start on the login bug.”
Write ·
memorix_project_contextSearches what past sessions saved about this project
Returns a short index of matching memories from any earlier session or agent, with filters for type, dates and status. It searches the current project unless you ask for all.
Try: “Search project memory for how we handled rate limiting.”
Read ·
memorix_searchOpens full notes found by a search
Fetches the full text of memories you pick by reference, including where each came from and what it links to.
Try: “Open the full notes for the two most relevant results.”
Read ·
memorix_detailSaves a decision, fix or gotcha for next time
Stores a note in the project's local memory, classified as a decision, fix, gotcha, change, discovery or similar, and indexes it for search.
Try: “Remember that we chose SQLite over Postgres because of offline use.”
Write ·
memorix_storeMarks old notes as done or archived
Marks memories as resolved or archived so they stop appearing in default searches. They can still be found with the all filter.
Try: “Mark the memory about the fixed login bug as resolved.”
Write ·
memorix_resolveBuilds a ready-to-use context bundle for a task
Combines relevant memories, code facts, freshness warnings, suggested files to read and verification hints into one bundle. Use it only to go beyond the project brief.
Try: “Build a context pack for refactoring the billing module, because the brief missed it.”
Read ·
memorix_context_packShows whether the code index is ready
Reports the state of the local code memory index for the current project and any external code-graph provider.
Try: “Is the code index up to date?”
Read ·
memorix_codegraph_statusStarts a work session and binds the project
Opens a new session, closes any previous one and returns a short card with the latest handoff. In a client that doesn't start in your repo, you pass the project folder here.
Try: “Start a session for the project in /Users/me/code/shop.”
Write ·
memorix_session_startImports and manages images, PDFs and audio for memory
Copies a file you name into Memorix's local media library, lists or shows assets, extracts PDF text and can attach an asset to a memory. Image and video generation stay off unless you enable them.
Try: “Import docs/architecture.png and attach it to the design notes.”
Write ·
memorix_media
Read from the source code
Permissions
medium risk · 6 things to watch
Medium risk
It writes notes about your work to local storage that other assistants read later, and its setup command can add hooks that capture prompts and file edits. Skip the setup command if you only want the server, and review stored notes for secrets.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It saves notes about your work, which later chats read
memorix_store and memorix_resolve write to the local memory, and other assistants on the same project read it back. Review what is stored if it could hold secrets. github.com
The setup command installs hooks and rewrites guidance files
memorix setup writes `memorix serve --mode lite` (21 tools) into the app's config and adds hooks that run `memorix hook` on prompts, tool calls, file edits and session end (for Claude Code and Codex: SessionStart, UserPromptSubmit, PostToolUse, PreCompact, Stop), plus guidance files such as CLAUDE.md and AGENTS.md. The events differ per app. Use --noHooks to skip the hooks, and preview removal with memorix uninstall --dry-run. github.com
Opening a project can build a local code index
memorix_project_context refreshes the Code Memory index when it is missing or stale, which writes snapshots to disk. Large repos take time. github.com
Importing media copies your files
memorix_media copies a file you name into a local media library outside the Git worktree. Image and video generation are off unless MEMORIX_MCP_MEDIA_GENERATION=1 is set. github.com
Larger tool sets add delete tools
Setting MEMORIX_MODE=full adds delete_entities, delete_observations and delete_relations, plus consolidate and deduplicate. The default micro set has none. github.com
The HTTP service stays on your machine by default
memorix background start serves on localhost port 3211. A non-local address needs MEMORIX_HTTP_AUTH_TOKEN or an explicit opt-out flag. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Talks to | None for the default stdio server; optional LLM, embedding and media providers you configure. Optional multi-device sync to GitHub, S3, filesystem or Postgres if you turn it on |
|---|---|
| Sign-in and keys | No account or API key. Notes are kept in a local SQLite database under the Git project and ~/.memorix. |
Questions
Is Memorix official?
Yes. The Memorix project ships this server for its own memory product. It is listed in the official MCP Registry as io.github.AVIDS2/memorix.
Which AI apps does it work with?
Any app that runs local MCP servers over stdio. The README also lists setup support for Claude Code, Codex CLI, GitHub Copilot CLI, Cursor, Windsurf, Gemini CLI, OpenCode, Kiro, Trae and others.
Do I need an API key?
No. The server runs on your machine without sign-in. Optional features such as model-based memory formation, embeddings and media generation use provider keys you set yourself.
Why does the config need "serve"?
Because running memorix with no arguments opens its terminal agent. MCP apps must start the server with memorix serve, or npx -y memorix serve.
How many tools does it show?
Nine by default (memorix serve). The lite set has 21, the team set 29 and the full set 48. Set MEMORIX_MODE (or the --mode flag) to change it. The memorix setup command writes --mode lite, so apps configured that way show 21.
Does it send my notes anywhere?
The default server keeps notes in local SQLite. Multi-device sync is off until you configure it. If you add an LLM or embedding provider, the text it processes goes to that provider.
Can I run it read-only?
No. Storing and resolving notes are part of the default tools, and the README has no read-only switch.
How do I uninstall it?
Run memorix uninstall --dry-run to preview, then memorix uninstall --yes --background --hooks to remove hooks and the background service. Remove the MCP config entry from your app by hand.
Information
updated Nov 2, 2026
- Publisher
- Official
- Package
memorix- Registry
- npm
- Transport
- stdio
- Auth
- None
- Version
- 1.10.2
- License
- Apache-2.0
- Last commit
- Oct 30, 2026
- Open issues
- 4
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 9Hide
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com
- github.com · github.com