Turbo Quant Memory — Claude Code MCP server: agent memory
A local store of decisions and notes for your coding agent. Search hands back short cards instead of whole documents, and nothing leaves your machine.
claude mcp add --env TQMEMORY_FTS_LANGUAGE=<TQMEMORY_FTS_LANGUAGE> --env TQMEMORY_EMBEDDING_BACKEND=<TQMEMORY_EMBEDDING_BACKEND> --env TQMEMORY_MIGRATE_ON_STARTUP=<TQMEMORY_MIGRATE_ON_STARTUP> --env TQMEMORY_SECRETS_PASSPHRASE=<TQMEMORY_SECRETS_PASSPHRASE> --transport stdio turbo-quant-memory -- uvx turbo-quant-memoryWhat Turbo Quant Memory does
The server gives an agent a persistent place to write while it works: decisions, lessons, patterns, session handoffs. It also keeps a compact index of your Markdown. A search returns cards of roughly 220 characters, and the agent pulls full content only when a card falls short. That is how it stops re-reading the same files and re-deriving the same conclusions.
Retrieval is hybrid: BM25 combined with dense vector search through RRF. There is a knowledge graph with lifecycle and linting. Cyrillic matches exactly out of the box, no extra setup. The package exposes 24 MCP tools, and server_info() reports the token savings it has accumulated, so you can check the numbers rather than take them on faith.
Python 3.11 or newer is required. On the first run fastembed pulls an embedding model of about 0.22 GB from Hugging Face, so that step needs a network connection; after it the server runs offline. There is no HTTP client and no telemetry in the package. Coming from 0.27.x or earlier, note that the distribution was renamed in 0.28.0, so uv tool upgrade turbo-memory-mcp no longer resolves — run uv tool install --force turbo-quant-memory once.
What it helps with
- keep session decisions after the context compacts
- find what the team decided about a topic and why
- stop re-reading the same files in a new session
- keep notes and code on your own disk
- search Cyrillic notes without extra configuration
Example prompts for the agent
- Index this project and record in memory why we picked this caching approach
- Search memory for what we decided about the migration scheme
- Show the token savings with
server_info()
How to install Turbo Quant Memory
Platform CLI
claude mcp add --env TQMEMORY_FTS_LANGUAGE=<TQMEMORY_FTS_LANGUAGE> --env TQMEMORY_EMBEDDING_BACKEND=<TQMEMORY_EMBEDDING_BACKEND> --env TQMEMORY_MIGRATE_ON_STARTUP=<TQMEMORY_MIGRATE_ON_STARTUP> --env TQMEMORY_SECRETS_PASSPHRASE=<TQMEMORY_SECRETS_PASSPHRASE> --transport stdio turbo-quant-memory -- uvx turbo-quant-memoryReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from turbo-quant-memory <command>
Platform CLI
codex mcp add turbo-quant-memory --env TQMEMORY_FTS_LANGUAGE=<TQMEMORY_FTS_LANGUAGE> --env TQMEMORY_EMBEDDING_BACKEND=<TQMEMORY_EMBEDDING_BACKEND> --env TQMEMORY_MIGRATE_ON_STARTUP=<TQMEMORY_MIGRATE_ON_STARTUP> --env TQMEMORY_SECRETS_PASSPHRASE=<TQMEMORY_SECRETS_PASSPHRASE> -- uvx turbo-quant-memoryReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from turbo-quant-memory <command>
Config file
{
"mcpServers": {
"turbo-quant-memory": {
"command": "uvx",
"args": [
"turbo-quant-memory"
],
"env": {
"TQMEMORY_FTS_LANGUAGE": "<TQMEMORY_FTS_LANGUAGE>",
"TQMEMORY_EMBEDDING_BACKEND": "<TQMEMORY_EMBEDDING_BACKEND>",
"TQMEMORY_MIGRATE_ON_STARTUP": "<TQMEMORY_MIGRATE_ON_STARTUP>",
"TQMEMORY_SECRETS_PASSPHRASE": "<TQMEMORY_SECRETS_PASSPHRASE>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from turbo-quant-memory <command>
Config file
{
"servers": {
"turbo-quant-memory": {
"type": "stdio",
"command": "uvx",
"args": [
"turbo-quant-memory"
],
"env": {
"TQMEMORY_FTS_LANGUAGE": "<TQMEMORY_FTS_LANGUAGE>",
"TQMEMORY_EMBEDDING_BACKEND": "<TQMEMORY_EMBEDDING_BACKEND>",
"TQMEMORY_MIGRATE_ON_STARTUP": "<TQMEMORY_MIGRATE_ON_STARTUP>",
"TQMEMORY_SECRETS_PASSPHRASE": "<TQMEMORY_SECRETS_PASSPHRASE>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from turbo-quant-memory <command>
Platform CLI
gemini mcp add -e TQMEMORY_FTS_LANGUAGE=<TQMEMORY_FTS_LANGUAGE> -e TQMEMORY_EMBEDDING_BACKEND=<TQMEMORY_EMBEDDING_BACKEND> -e TQMEMORY_MIGRATE_ON_STARTUP=<TQMEMORY_MIGRATE_ON_STARTUP> -e TQMEMORY_SECRETS_PASSPHRASE=<TQMEMORY_SECRETS_PASSPHRASE> turbo-quant-memory uvx turbo-quant-memoryReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from turbo-quant-memory <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 Turbo Quant Memory
How do I install Turbo Quant Memory in Claude Code?
Install the package first: uv tool install turbo-quant-memory. Then register the server: claude mcp add turbo-quant-memory -e TQMEMORY_SECRETS_PASSPHRASE=<TQMEMORY_SECRETS_PASSPHRASE> -- uvx turbo-quant-memory.
What does Turbo Quant Memory do?
It is an MCP server that gives a coding agent local memory and a knowledge graph. It stores decisions, lessons and session handoffs, and search returns compact cards rather than full documents.
What does the server need to run?
Python 3.11 or newer. On the first run fastembed downloads an embedding model of about 0.22 GB from Hugging Face, after which the server works offline.
