NeuroStack — Claude Code MCP server: features and setup
NeuroStack indexes Markdown notes for MCP search, link traversal, and session memories. Use it to give an AI client access to an existing notes vault.
claude mcp add --transport http neurostack https://mcp.neurostack.sh/mcpWhat NeuroStack does
NeuroStack builds a SQLite index from .md files, combining text search, embeddings, and a graph of wiki links. An MCP client can retrieve ranked evidence with note paths, while the AI handles the reasoning. Indexing leaves the Markdown files unchanged.
The server can flag notes that have gone stale and turn decisions from AI sessions into saved memories. For Claude Code, connect through the hosted HTTP endpoint or run the package locally with npx and NEUROSTACK_VAULT_ROOT. These are separate connection options; choose one configuration method.
Lite offers keyword search and link-based connections without GPU or Ollama. Full adds semantic search, generated summaries, and topic grouping through Ollama. The project recommends a GPU or a CPU with 6+ cores for Full.
What it helps with
- search a Markdown vault for project context
- follow wiki-link connections between notes
- spot notes that may have become stale
- save session decisions as AI memories
Example prompts for the agent
- Search my notes for decisions about the project architecture.
- Check the vault for notes that may be out of date.
- Save the decisions and root causes from this session as memories.
How to install NeuroStack
1Platform CLI: remote server
claude mcp add --transport http neurostack https://mcp.neurostack.sh/mcp2Platform CLI
claude mcp add --env NEUROSTACK_VAULT_ROOT=<NEUROSTACK_VAULT_ROOT> --transport stdio neurostack -- npx -y neurostackReplace <…> with your own values
1Platform CLI: remote server
codex mcp add neurostack --url https://mcp.neurostack.sh/mcp2Platform CLI
codex mcp add neurostack --env NEUROSTACK_VAULT_ROOT=<NEUROSTACK_VAULT_ROOT> -- npx -y neurostackReplace <…> with your own values
Config file
{
"mcpServers": {
"neurostack": {
"url": "https://mcp.neurostack.sh/mcp"
}
}
}{
"mcpServers": {
"neurostack": {
"command": "npx",
"args": [
"-y",
"neurostack"
],
"env": {
"NEUROSTACK_VAULT_ROOT": "<NEUROSTACK_VAULT_ROOT>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"neurostack": {
"type": "http",
"url": "https://mcp.neurostack.sh/mcp"
}
}
}{
"servers": {
"neurostack": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"neurostack"
],
"env": {
"NEUROSTACK_VAULT_ROOT": "<NEUROSTACK_VAULT_ROOT>"
}
}
}
}Replace <…> with your own values
1Platform CLI: remote server
gemini mcp add -t http neurostack https://mcp.neurostack.sh/mcp2Platform CLI
gemini mcp add -e NEUROSTACK_VAULT_ROOT=<NEUROSTACK_VAULT_ROOT> neurostack npx -y neurostackReplace <…> with your own values
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 NeuroStack
How do I install NeuroStack in Claude Code?
For the HTTP connection, run claude mcp add --transport http neurostack https://mcp.neurostack.sh/mcp.
What does NeuroStack do?
It indexes Markdown notes in SQLite and exposes search and wiki-link relationships to MCP clients. It can also save memories from AI sessions.
What is the difference between Lite and Full mode?
Lite provides keyword search, note links, and stale-note checks without GPU or Ollama. Full adds semantic search, summaries, and topic grouping using Ollama.
