mnemo — Claude Code MCP server: what it does, how to install
Persistent memory engine for coding agents using SQLite, hybrid retrieval, and knowledge graphs. Keeps context across assistant sessions.
claude mcp add mnemo -e API_KEYS=<API_KEYS> -- uvx mnemo-mcpWhat mnemo does
Mnemo maintains assistant memory inside a local SQLite file operating in WAL mode. To pull relevant context, it merges FTS5 text matches with sqlite-vec embeddings using Reciprocal Rank Fusion, then reranks candidates while factoring in temporal decay and importance weights. The v2 update adds a temporal knowledge graph with bitemporal validity columns to inspect past states.
The tool runs as mnemo-mcp via uvx. You need to pass the API_KEYS environment variable at startup, because model reranking cells fail without valid credentials. Embedded cloud synchronisation was removed in earlier revisions, so routine backups rely on standard tools like rclone.
What it helps with
- retain project context across separate agent sessions
- retrieve historical decisions via combined text and vector retrieval
- track changing relationships between entities over time
- suppress outdated technical context with recency scoring
Example prompts for the agent
- Query mnemo memory to see what database schema choices were finalized
- Record this architectural decision in permanent memory with high importance
- Inspect the entity history and graph connections for the auth service
How to install mnemo
Platform CLI
claude mcp add mnemo -e API_KEYS=<API_KEYS> -- uvx mnemo-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mnemo-mcp <command>
Platform CLI
codex mcp add mnemo --env API_KEYS=<API_KEYS> -- uvx mnemo-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mnemo-mcp <command>
Config file
{
"mcpServers": {
"mnemo": {
"command": "uvx",
"args": [
"mnemo-mcp"
],
"env": {
"API_KEYS": "<API_KEYS>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mnemo-mcp <command>
Config file
{
"servers": {
"mnemo": {
"type": "stdio",
"command": "uvx",
"args": [
"mnemo-mcp"
],
"env": {
"API_KEYS": "<API_KEYS>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mnemo-mcp <command>
Platform CLI
gemini mcp add -e API_KEYS=<API_KEYS> mnemo uvx mnemo-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mnemo-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 mnemo
How to install mnemo in Claude Code?
Run claude mcp add mnemo -e API_KEYS=<API_KEYS> -- uvx mnemo-mcp with your actual API key. Alternatively, pass transport explicitly: claude mcp add --env API_KEYS=<API_KEYS> --transport stdio mnemo -- uvx mnemo-mcp.
How does mnemo perform hybrid search?
It evaluates both SQLite FTS5 keyword queries and sqlite-vec vector similarities, fusions them via RRF, and refines final rankings through a cross-encoder model.
Where does mnemo store its records?
All captured context stays stored locally inside a SQLite database using write-ahead logging.
