obsidian-mcp-plugin: turn your shared Obsidian vault into a team brain for any AI agent

Short version: If your team shares one Obsidian vault and you want Claude, Cursor, Gemini, and any other MCP agent to read and write that vault concurrently, obsidian-mcp-plugin is the right tool. It exposes your vault over HTTP so multiple agents and multiple people can operate on the same brain. 10-minute install. By aaronsb, MIT licensed.

What is obsidian-mcp-plugin?

obsidian-mcp-plugin is a community Obsidian plugin that starts an HTTP server inside Obsidian and exposes 40+ semantic tools (search, template expansion, structural edits) to any AI agent that speaks the Model Context Protocol. It's the serious version of the Obsidian-to-AI bridge — built for teams, multi-agent setups, and anyone who has outgrown single-user plugins.

Where obsidian-mcp-tools (the simpler cousin) runs over stdio and is tuned for one Obsidian + one Claude, obsidian-mcp-plugin runs as an HTTP service that any number of clients can hit. That shift matters when you have 10+ people writing to the same vault, or when you want Claude and Cursor both reading your team's playbook.

Who this is for

Teams with 10+ people sharing one vault of SOPs, playbooks, meeting notes, and decisions.
Setups where Claude Code, Cursor, and a custom backend agent all need the same brain.
Power users who want semantic search over notes, not just filename matching.
Anyone running multiple agent clients on the same computer.

Skip this if

You're a solo founder with a personal vault. obsidian-mcp-tools (stdio) or claudian (in-app) are simpler and do everything you need.

What problem it solves

Most AI-vault bridges assume one person with one agent. As soon as you add a second person or a second client, the stdio model breaks down — every client wants its own process, and they fight over vault access.

obsidian-mcp-plugin fixes that by running a single HTTP server inside Obsidian. Every agent — Claude Code, Cursor, Gemini, your custom bot — connects to the same URL and gets a consistent view of the vault. Team-shared playbooks, collaborative SOPs, and multi-agent workflows become possible without each client reinventing the bridge.

How to install it (plain English)

Install the plugin. Obsidian Settings → Community plugins → Browse → search "MCP plugin" by aaronsb → Install and Enable.
Start the server. Settings → MCP Plugin → toggle "Start server". It listens on a local port (usually 27124). The plugin shows the exact URL.
Point your agent at the HTTP endpoint. In your client's MCP config, add an HTTP transport entry with the URL the plugin displays. Claude Code and recent MCP clients support HTTP transport directly.
Restart your agent and test. Start a new chat. Ask "list the notes in my vault" — you should see real titles.

For team use across machines, you'll need to expose the port via a tunnel (Cloudflared, ngrok, Tailscale) or run Obsidian on a shared machine. Full walkthrough: /memory/tools/obsidian-mcp-plugin.

What you can do with it (for a non-technical founder)

"Our onboarding playbook" shared across Claude + Cursor + ChatGPT — all three agents reference the same source of truth.
Semantic search — "find notes similar to our Q3 brand guide" pulls thematic matches, not just keyword hits.
Template-driven note creation — "create a new customer-interview note using our standard template" works from any agent.
Multi-person concurrent use — your CEO and ops lead can both have AI agents reading the vault at the same time without conflict.
One brain, many surfaces — add a Slack bot or email bot that reads the same vault via HTTP.

What CLO adds on top

obsidian-mcp-plugin gives your agents read/write access to the shared brain. Cognition CLO models what each person on the team is retaining from that brain. The plugin means your agents know your SOP; CLO tells you that 3 of your 12 team members haven't touched it in 30 days and are likely to miss a key step. Different layers, both necessary.

FAQ

What's the difference between this and obsidian-mcp-tools?

obsidian-mcp-tools uses stdio transport — simpler, one-to-one. obsidian-mcp-plugin uses HTTP — scales to multiple clients and multiple users. Use tools for solo, plugin for teams.

Can I expose the server over the internet?

Possible, but think about it carefully. The plugin is designed for local/LAN use. If you tunnel it publicly, put auth in front (Cloudflare Access, Tailscale ACLs). Don't just open the port to the internet.

Does it support multiple agent clients on the same computer?

Yes. Claude Code, Cursor, and any other MCP-HTTP client can all hit the same local URL. They share the vault's state.

What's the performance like?

For typical vaults (a few thousand notes), HTTP overhead is negligible. Semantic search is the slowest operation; still sub-second in most cases.

Does it work with Obsidian Sync?

Yes. The plugin operates on whatever Obsidian shows it. If your sync solution keeps the vault up to date, the plugin sees the latest notes.

Ready to install? Full walkthrough at /memory/tools/obsidian-mcp-plugin. Credit to @aaronsb for building and maintaining it — star the repo if it saves your team time.