Your Obsidian vault remembers what matters.

Every other Obsidian MCP server is basically grep with extra steps — keyword search, sort by modified date, and hope the LLM picks the right file from 50 matches. None of them know which notes are important to you.

Hebbian Vault is different: it has memory. Every time you read a note, its rank strengthens. Unused notes fade. Hub pages (the ones with lots of incoming wikilinks) surface first via PageRank. Keyword matches ride on top via BM25. All four signals — usage, centrality, keyword relevance, recency — merge through Reciprocal Rank Fusion into a single ranking that gets smarter the more you use your vault.

Seven tools, working out of the box on any Obsidian vault — no plugin required, no Obsidian app running, no cloud roundtrip for local use:

vault_search — hybrid-ranked search that learns from your retrieval patterns
vault_read — fetch a note by path; increments its retrieval count automatically
vault_neighbors — find structurally adjacent notes via Personalized PageRank
vault_hot — top-ranked notes right now (what your vault thinks matters)
vault_stats — vault size, link density, top tags, PageRank distribution
vault_health — broken wikilinks, orphans, missing frontmatter
configure_vault — point the server at a new vault any time

Why it matters. When your AI has access to a vault that ranks by actual importance, answers stop being "here are 20 possibly-relevant files" and start being "here is the hub note you actually wrote about this, and its three most-linked neighbors." The signal was always in the graph structure and your own usage history — Hebbian Vault just surfaces it.

Works with any vault from day one. Starts cold (uniform ranks), warms up as you use it. No training data required — the vault itself is the training signal. Hebbian learning from neuroscience, applied to information retrieval: the pathways you walk most, strengthen.