Genome
Local, offline agent memory with no LLM calls on write and an auditable record
- Pricing
- Open source
- Platforms
- CLI, self-hosted, macOS
- Runs
- Locally
- Version
- 1.3.2
- Listing updated
- today
Similar MCP servers
About Genome
Genome is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.
Genome's bet is that an agent does not need a language model to decide what to remember. Each message is embedded locally with all-MiniLM-L6-v2 and stored in SQLite or Postgres, so a write takes about 10 ms with no network and no LLM call. Reads use exact cosine search within a tenant's rows, with an optional local cross-encoder reranker. Because embedding is deterministic, replaying the same input reproduces the same store, which the project presents as the basis for auditability.
An opt-in bi-temporal layer records each fact at the time it became true, so a question such as what someone's city was in March can have a different correct answer than what it is now. That step does use one LLM call for belief extraction.
The README is open about its limits: it says accuracy ties Mem0 on LoCoMo and LongMemEval rather than beating it, and the claimed advantages are cost, speed, offline use and the audit record. The MCP server stores memories in ~/.genome/memories.db and gives the agent four tools. A genome.verify command demonstrates the offline write path with outbound sockets blocked.
Install
Run in a terminal
claude mcp add genome -- uvx genome-memoryAdd to claude_desktop_config.json
{
"mcpServers": {
"genome": {
"command": "uvx",
"args": [
"genome-memory"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"genome": {
"command": "uvx",
"args": [
"genome-memory"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"genome": {
"type": "stdio",
"command": "uvx",
"args": [
"genome-memory"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"genome": {
"command": "uvx",
"args": [
"genome-memory"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"genome": {
"command": "uvx",
"args": [
"genome-memory"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Tools this server provides (4)
Taken from the project documentation; the running server may differ.
- remember
- Store a memory
- recall
- Retrieve relevant memories
- forget
- Remove a memory
- reset_memories
- Clear all stored memories
What it does
- No-LLM writes
- Messages are embedded locally and stored in about 10 ms with zero model calls.
- Fully offline
- The write path works with outbound network blocked.
- Deterministic store
- The same input reproduces the same vectors and content.
- Point-in-time facts
- Opt-in bi-temporal layer answers what was true at a given date.
- Local reranking
- An optional cross-encoder reranker improves recall on your machine.
- Self-check command
- `python -m genome.verify` prints a pass/fail receipt for offline operation.
Good for
- Give Claude Code persistent memory across sessions with no cloud account
- Run agent memory in an air-gapped environment
- Keep an auditable record of what an agent stored
- Ask what was true about a person at an earlier date
- Compare memory cost against LLM-extraction systems
Pricing
Open source
Free; open source with a licence file in the repository
Genome FAQ
Does it need an API key?
The MCP server needs no key and keeps data on your machine.
Where are memories stored?
In ~/.genome/memories.db by default, or Postgres if configured.
Is it more accurate than Mem0?
No claim of that. The README says answer accuracy ties Mem0 on the benchmarks it ran.
How do I install it?
pip install "genome-memory[mcp]", then add the genome-mcp command to your client config.
Is Genome free?
Yes. Genome is open source, so you can use and self-host it for free.
How do I add Genome to Claude Code?
Run this command in your terminal: claude mcp add genome -- uvx genome-memory
What are the best alternatives to Genome?
The closest alternatives to Genome are Codebase Memory, XERJ, headroom, omniget and next-ai-draw-io.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.



