Neuromcp
Local persistent memory for AI agents with hybrid search and a wiki knowledge base
- Pricing
- Open source
- Platforms
- CLI, macOS, Linux
- Runs
- Locally
- Version
- 0.30.2
- Listing updated
- today
Similar MCP servers
About Neuromcp
Neuromcp is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.
Neuromcp keeps an assistant's memory on your own machine instead of a cloud service. Memories live in a single SQLite file, and recall blends vector search using the nomic-embed-text model, BM25 full-text matching, graph links and a learned usefulness prior. A built-in browser shows entities and their relations as a graph served from localhost.
A second layer is a Markdown wiki, git-tracked, that is injected into a session at startup, updated during work and committed when the session ends. A consolidation pipeline scheduled through launchd rolls raw session logs into wiki pages. Setup is one command, npx neuromcp-init, with npx neuromcp-doctor to diagnose problems.
The README reports 96.08% on a 102-question LongMemEval-S sample and 93.3% recall at 5 with 500 distractors, and openly warns about sample size. It states there are no API keys, no telemetry and no cloud sync.
Install
Run in a terminal
claude mcp add neuromcp -- npx -y neuromcpAdd to claude_desktop_config.json
{
"mcpServers": {
"neuromcp": {
"command": "npx",
"args": [
"-y",
"neuromcp"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"neuromcp": {
"command": "npx",
"args": [
"-y",
"neuromcp"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"neuromcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"neuromcp"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"neuromcp": {
"command": "npx",
"args": [
"-y",
"neuromcp"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"neuromcp": {
"command": "npx",
"args": [
"-y",
"neuromcp"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
What it does
- Local SQLite memory
- Stores everything in one file on your own disk.
- Hybrid retrieval
- Combines vector, full-text, graph and usefulness signals.
- Wiki knowledge base
- Compiles git-tracked Markdown knowledge injected at session start.
- Memory browser
- Shows a draggable entity graph served from localhost.
- Automatic consolidation
- Periodically merges session logs into wiki pages.
- One-command setup
- Detects installed MCP clients and writes configs.
Good for
- Give a coding agent persistent project knowledge
- Carry memory across Claude, Cursor and other MCP clients
- Keep conversation history private and local
- Inspect what an agent has remembered
Pricing
Open source
Open source; no API keys or subscription
Neuromcp FAQ
Does data leave my machine?
The README says no data leaves your laptop and there is no telemetry or cloud sync.
Which clients work with it?
Any MCP client, including Claude Desktop, Cursor, Windsurf, Codex CLI, Continue, LibreChat and Open WebUI.
How good is recall?
The README reports 96.08% on a 102-question LongMemEval-S sample, with sample-size caveats.
How do I set it up?
Run npx neuromcp-init, or run the bare server with npx neuromcp.
Is Neuromcp free?
Yes. Neuromcp is open source, so you can use and self-host it for free.
How do I add Neuromcp to Claude Code?
Run this command in your terminal: claude mcp add neuromcp -- npx -y neuromcp
What are the best alternatives to Neuromcp?
The closest alternatives to Neuromcp 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.



