Launched this week
Seekstone
Give Claude your Obsidian vault — without the context tax
7 followers
Give Claude your Obsidian vault — without the context tax
7 followers
Seekstone is an open-source MCP server that connects Claude (or any MCP client) straight to your Obsidian vault by reading it from disk. No plugin, no app running, nothing leaves your machine. Search returns ~2 KB of ranked excerpts — not megabytes of full notes. Reproducible benchmarks included.
The local-only approach is honestly what sold me on this, super clean setup. One thing I'd love to see is fuzzy or typo-tolerant search, because right now if I slightly misspell a tag or a term in my vault the results just miss it entirely. Even a simple edit-distance fallback on titles would make this feel way more forgiving day to day.
@mustafasobgmyg Semantic type search is on our roadmap to implement. Appreciate the feedback, and this is something we will prioritize on the roadmap. Ideally, finding relevant information is the tool's #1 job.
@mustafasobgmyg Coming back to correct my own earlier reply, because I went and checked the code — typo tolerance is actually already in there, on by default. Every search_vault query runs with edit-distance fuzzy matching plus prefix matching, across titles, body text, and tags (tags are even boosted). "benchamrk" still finds "benchmark" — roughly one typo's worth of slack on short words, two on longer ones.
So if a slightly misspelled term came back empty, I'd genuinely love to see the exact query. Two likely culprits: the typo exceeded the edit-distance budget, or it went through one of the structured tools (exact tag filters don't fuzz) — both are fixable, and I'll turn your example into a test case.
What I meant by the roadmap comment is the different beast: semantic search, matching by meaning ("car" → "automobile"). That's still ahead of us — but for misspellings, you're covered today.
honestly this is exactly what i needed for my vault, love that nothing leaves my machine. one thing that would make it even better is supporting some kind of incremental indexing so the initial setup on a large vault doesnt take forever, maybe with progress feedback so you know its not just hanging
@ebraryqku Appreciate the feedback. Makes perfect sense. We will include that into the roadmap for consideration!