Capture and promote are familiar; drift and blind-spot detection are not. Looking at which tags gained and lost weight across a year is a different question from searching your notes, and it needs the graph that vault_analyze builds first. Migration is the other thing done carefully here — an existing vault is classified and converted with the originals archived rather than overwritten.
A knowledge-vault server organised in three layers — capture, connect, reflect. It writes into a local directory in an Obsidian-compatible format, and can migrate an existing vault into that shape.
- vault_capture saves a refined thought with automatic tagging, and deletes captures you no longer want
- vault_read searches the vault, reads a file, and lists captures by status
- vault_promote turns a capture into a structured note with wikilinks inserted automatically
- vault_analyze rebuilds the knowledge graph and runs Louvain clustering, N-degree connection queries and orphan detection
- vault_topic gathers material, then creates or updates a map-of-content style topic note
- vault_reflect produces three views: a snapshot of the current knowledge landscape, interest drift over time, and blind-spot and bridge detection
- vault_init seeds an empty vault from a template, or migrates an existing Obsidian vault — classifying notes, converting todos and archiving the originals under _archive/
- Tags are drawn from a tags.yaml at the vault root first, then from existing notes' frontmatter, then from a built-in list of default domains
Uv, with the server run as `uvx obsidian-in-a-vat-mcp` — package version 0.3.12 — or from the container image ghcr.io/oliverxuzy-ai/obsidian-in-a-vat. VAULT_LOCAL_PATH points at your vault directory; an empty directory is fine, since the first initialisation creates the structure. No account.
One command — uvx obsidian-in-a-vat-mcp