Build token-efficient AI memory from your Markdown vault using tiered search and GraphRAG. For developers integrating persistent, semantic knowledge into Claude or other LLMs without breaking context budgets.
io.github.raphasouthall/neurostack
https://mcp.neurostack.sh/mcp
HTTP
No auth required
How models use it and what it is built for.
Build token-efficient AI memory from your Markdown vault using tiered search and GraphRAG. For developers integrating persistent, semantic knowledge into Claude or other LLMs without breaking context budgets.
Hosted endpoint — paste into any MCP client.
Configuration this server reads at startup.
Path to your Markdown vault (e.g. ~/brain, ~/notes)
Where to find authoritative docs and source for NeuroStack.
Paste any of these into Agent Studio after connecting NeuroStack.
Common questions about connecting and running NeuroStack.
What format should my Markdown vault be in?
Point NEUROSTACK_VAULT_ROOT to any directory containing Markdown files (e.g., ~/brain or ~/notes). The server indexes all .md files recursively and builds a searchable knowledge graph from their content.
How does this reduce token usage compared to dumping all notes into context?
Tiered search and GraphRAG retrieve only the most relevant notes semantically, rather than sending your entire vault. This keeps context windows lean while maintaining access to your full knowledge base.
Do I need to set up authentication or API keys?
No API keys are required. You only need to set NEUROSTACK_VAULT_ROOT to your local Markdown directory and connect to the hosted MCP endpoint at https://mcp.neurostack.sh/mcp.
Can I use this with Claude or only specific LLMs?
As an MCP server, it works with any LLM client that supports the Model Context Protocol, including Claude. Check your client's MCP integration docs for setup steps.
What happens if my vault has thousands of notes?
GraphRAG and tiered search are designed to scale efficiently. Start with npx neurostack@0.12.0 and monitor token usage; the server prioritizes high-relevance results to keep overhead minimal even on large vaults.
Connect NeuroStack to Claude, GPT, Gemini, DeepSeek and 30+ AI models in MCP Agent Studio. Compare answers side-by-side, save reusable agent presets, share runs — all in your browser, no install required.
Open Agent Studio