Magnemo — Claude Code MCP server: what it does, how to install
Stores agent memory in local markdown files under your supervision. It makes zero network calls and writes facts only after you approve them.
claude mcp add --env MAGNEMO_VAULT=<MAGNEMO_VAULT> --env MAGNEMO_AGENT=<MAGNEMO_AGENT> --transport stdio magnemo -- uvx magnemo-mcpWhat Magnemo does
Magnemo keeps agent notes on your hard drive in a folder of markdown files. When an assistant learns a project detail, the entry sits in staging until you confirm it in chat. Nothing saves without your explicit consent. Each record logs both the creator and the reviewer in file headers and an append-only ledger.
The server runs without external model requests or vector embeddings. It relies on keyword search, so semantic matching is absent. Python 3.11 or newer is required to execute the tool. Deleted items are never discarded outright: rejected ideas keep refusal notes, while superseded entries move into an archive with pointers to current versions.
What it helps with
- keep project context in local markdown files
- approve or reject agent memory entries directly in chat
- inspect author and approver details for every saved fact
- maintain long-term memory without making external API calls
Example prompts for the agent
- Use Magnemo to remember what this project is
- Check what you remember about our database setup
- Store the deployment procedure in Magnemo
How to install Magnemo
Platform CLI
claude mcp add --env MAGNEMO_VAULT=<MAGNEMO_VAULT> --env MAGNEMO_AGENT=<MAGNEMO_AGENT> --transport stdio magnemo -- uvx magnemo-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from magnemo-mcp <command>
Platform CLI
codex mcp add magnemo --env MAGNEMO_VAULT=<MAGNEMO_VAULT> --env MAGNEMO_AGENT=<MAGNEMO_AGENT> -- uvx magnemo-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from magnemo-mcp <command>
Config file
{
"mcpServers": {
"magnemo": {
"command": "uvx",
"args": [
"magnemo-mcp"
],
"env": {
"MAGNEMO_VAULT": "<MAGNEMO_VAULT>",
"MAGNEMO_AGENT": "<MAGNEMO_AGENT>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from magnemo-mcp <command>
Config file
{
"servers": {
"magnemo": {
"type": "stdio",
"command": "uvx",
"args": [
"magnemo-mcp"
],
"env": {
"MAGNEMO_VAULT": "<MAGNEMO_VAULT>",
"MAGNEMO_AGENT": "<MAGNEMO_AGENT>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from magnemo-mcp <command>
Platform CLI
gemini mcp add -e MAGNEMO_VAULT=<MAGNEMO_VAULT> -e MAGNEMO_AGENT=<MAGNEMO_AGENT> magnemo uvx magnemo-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from magnemo-mcp <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Magnemo
How to install Magnemo in Claude Code?
Run claude mcp add magnemo -e MAGNEMO_VAULT=<MAGNEMO_VAULT> -- uvx magnemo-mcp. Replace <MAGNEMO_VAULT> with the absolute path to your initialized vault directory.
What does Magnemo do?
It provides persistent memory for AI agents using local markdown files. Every write requires human approval in the conversation before it enters the storage vault.
Does Magnemo require an embedding API?
No third-party APIs or embedding models are used. The server executes entirely on your device and locates information using keyword matching.
