genome — Claude Code MCP server: what it does, how to install
Local memory for AI agents: writes with no LLM calls, semantic recall, offline operation and an auditable fact history.
claude mcp add genome -- uvx genome-memoryWhat genome does
genome keeps messages in a local store — SQLite or Postgres — and searches them by cosine similarity. The write path makes no LLM calls and touches no network: text goes through the local all-MiniLM-L6-v2 embedder, and the same text always produces the same vector. Replay a journal and you get the same store back, which is what makes the record checkable instead of trusting whatever a model once decided mattered.
An optional bi-temporal layer files each fact under the moment it became true in the world rather than the moment it arrived. That is how point-in-time questions stay answerable — older values are not overwritten by newer ones. Belief extraction is the only step that calls an LLM, and it is opt-in.
The first run pulls the embedding model (~90 MB), so a cold machine sits quiet for a while. You can check the cost and offline claims without any API key: python -m genome.verify writes memories with outbound sockets blocked and prints a pass/fail receipt. Accuracy parity with Mem0 is a separate, larger check and needs an LLM key.
What it helps with
- carry memory across sessions without sending data off the machine
- answer what a fact was on a given date, not just what it is now
- re-derive the stored record and audit what is in it
- keep LLM calls and network traffic out of the write path
- hold several tenants' memories in one store
Example prompts for the agent
- Remember that the client moved to Austin in February 2025
- What was the client's city in March 2024?
- List every fact about Priya that was valid in May 2023
How to install genome
Platform CLI
claude mcp add genome -- uvx genome-memoryuvx runs the command named after the package; if the README names another one, use uvx --from genome-memory <command>
Platform CLI
codex mcp add genome -- uvx genome-memoryuvx runs the command named after the package; if the README names another one, use uvx --from genome-memory <command>
Config file
{
"mcpServers": {
"genome": {
"command": "uvx",
"args": [
"genome-memory"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from genome-memory <command>
Config file
{
"servers": {
"genome": {
"type": "stdio",
"command": "uvx",
"args": [
"genome-memory"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from genome-memory <command>
Platform CLI
gemini mcp add genome uvx genome-memoryuvx runs the command named after the package; if the README names another one, use uvx --from genome-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 genome
How do I install genome in Claude Code?
With one command: claude mcp add genome -- uvx genome-memory.
What does genome do?
It is a local memory MCP server for agents. Messages are embedded on the spot, stored in SQLite or Postgres, and handed back as context on request — no API key, nothing leaving your machine.
Does genome need an API key?
Not for storing or searching. An LLM key is only needed for belief extraction and for the head-to-head accuracy check against Mem0.
