Obsidian PKM MCP Knowledge management
If you use Obsidian as your personal knowledge base, you've probably noticed the friction: Claude is your thinking partner, but it can't see your notes. You copy-paste relevant sections into each conversation. You maintain two systems that don't talk to each other. The knowledge you've built up in Obsidian sits next to Claude, not inside it.
There are local Obsidian MCP plugins that solve this — but they require running a local server, staying on the same machine, and exposing your entire vault to the connection. For people who want Claude to access their knowledge base on any device without spinning up infrastructure, that's more overhead than it's worth.
Stash is a different approach: a hosted MCP server where you store the knowledge Claude actually needs — selectively, without your whole vault — and it's available from any Claude session, on any device.
| Approach | Setup | Access | Privacy |
|---|---|---|---|
| Local Obsidian MCP plugin | Run local server, configure port, keep running | Same machine only | Full vault exposed to connection |
| Paste from Obsidian | None | Any device | You control what you share |
| Stash | 30 seconds (paste a URL) | Any device, any Claude session | You choose what to store |
The trade-off with Stash is intentional: you don't get automatic full-vault sync. You store what you want Claude to know — which is usually a small, curated subset of your vault, not 3,000 notes about everything you've ever read.
I'm writing an essay on attention and deep work. What have I already stashed about this?
You have 8 records in 'deep-work'. Key themes across them: you're skeptical of the "flow state" framing (three notes flag it as too mystical); you find the Cal Newport framework useful but too dismissive of reactive work; your best writing sessions have been 90 minutes, not 4-hour blocks. You've flagged a quote from a 2023 Simone Weil essay you want to use — want me to pull it?
Yes. And what were my notes on the Huberman attention protocol?
From your Huberman record: you logged the visual focus drill (gaze on a fixed point for 30–60s before deep work), noted it works for you but only if the session starts within 10 minutes of the drill, and flagged skepticism about the cold exposure claim. You also noted: "feels like a ritual more than a mechanism, but rituals work."
https://app.stashlite.com/mcp → Save
The fastest migration: open Obsidian, find the 10–15 notes you've referred to most in the last month, and have Claude stash them. You can do this in a single session — paste each note, ask Claude to stash it with a short title and collection name.
After one session, Claude will have access to your most-used knowledge without you having ever configured a local server or exposed your full vault.
Stash is designed for fast, cheap retrieval. When Claude calls search() on your records, it returns matching records — not your entire knowledge base. For a 500-record search, Stash uses around 192 tokens. A system that loads full Obsidian context for every query would use orders of magnitude more.
This matters because token cost is real. Every token Claude reads competes for context. Stash returns what's relevant, not everything you've ever written.
The free tier gives you 2,500 records and 50 queries per month. For a personal knowledge base with curated notes, most people never hit the free limit. Pro is £8/month for 100,000 records and 1,000 queries — appropriate if you're migrating a large active collection or querying heavily.
https://app.stashlite.com/mcp
Stash doesn't replace Obsidian. It doesn't do backlinks, graph view, plugins, or long-form note taking. Obsidian is your thinking environment. Stash is the bridge between that thinking environment and Claude — the curated layer of knowledge that makes Claude useful across sessions without requiring you to paste your vault into every conversation.