neurodivergent-memory — MCP server for agent memory graphs
A local MCP server that keeps an agent's notes as a knowledge graph split into five districts, with BM25 ranking. Useful when context keeps vanishing between sessions.
claude mcp add neurodivergent-memory -- npx -y neurodivergent-memoryWhat neurodivergent-memory does
The server is written in TypeScript and speaks over stdio. Two layers sit inside: a memory graph with bidirectional edges and a BM25 index that ranks search hits. Everything persists to a local JSON snapshot, so memory survives a restart and no cloud model is involved.
Memories live in five districts: logical_analysis, emotional_processing, practical_execution, vigilant_monitoring, creative_synthesis. A node carries tags, emotional valence and intensity, a project id and an epistemic status. The tools cover the usual loop: store_memory, retrieve_memory, update_memory, delete_memory, connect_memories. For lookup there is search_memories with filters and a recency bias, graph walking via traverse_from, similarity via related_to, and memory_stats for totals.
Node.js 24 is required. The quick start in the README sets Node up through Chocolatey on Windows or nvm on Linux and macOS, then runs npx neurodivergent-memory@latest init-agent-kit. The snapshot and WAL are plain files on disk, and storage_diagnostics reports which path is in use. Point the snapshot at a directory the process cannot write to and nothing is kept between sessions.
Some tools mirror prompts for clients that call tools but skip prompts: prepare_memory_city_context, prepare_synthesis_context, prepare_packetized_synthesis_context.
What it helps with
- carry agent notes across sessions
- find a memory by meaning with BM25
- link two notes with a bidirectional edge
- sort notes into cognitive districts
- count memories, orphans and per-project totals
Example prompts for the agent
- Store the caching decision in memory and tag it
redis - Search memory for everything about the DB migration and show the connected nodes
- Run a synthesis over the notes of project
billingfrom this week
How to install neurodivergent-memory
Platform CLI
claude mcp add neurodivergent-memory -- npx -y neurodivergent-memoryPlatform CLI
codex mcp add neurodivergent-memory -- npx -y neurodivergent-memoryConfig file
{
"mcpServers": {
"neurodivergent-memory": {
"command": "npx",
"args": [
"-y",
"neurodivergent-memory"
]
}
}
}Config file
{
"servers": {
"neurodivergent-memory": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"neurodivergent-memory"
]
}
}
}Platform CLI
gemini mcp add neurodivergent-memory npx -y neurodivergent-memoryStep 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 neurodivergent-memory
How do I install neurodivergent-memory in Claude Code?
Run claude mcp add neurodivergent-memory -- npx -y neurodivergent-memory.
What does neurodivergent-memory do?
It stores notes as a knowledge graph and exposes them over MCP: writing, search, graph traversal, stats. Search is ranked with BM25 and the data sits in a local snapshot.
Does it need a cloud LLM?
No. The server runs locally over stdio and never calls an outside model. You need Node.js 24 to start it.
