364 MCP Servers Can't Fix Obsidian's Real Problem

Obsidian has 1.5 million users, 2,700 plugins, and 22% year-over-year growth. It deserves every bit of that momentum. So why did I spend a year building a competing product?

Because Obsidian gives you a workspace. Most people need a system that works without them.

The vault problem

Every Obsidian power user I know has hit the same wall. You spend a weekend setting up your vault. Folders, templates, daily notes, dataview queries, the works. It feels incredible for about three weeks. Then life gets busy.

A Reddit user put it plainly: "My Obsidian vault turned into a junk drawer once life got busy." Another: "The 'maintaining the system' part is what kills most setups for me too."

This is not a criticism of Obsidian. It is a criticism of the architecture that treats you as the permanent maintenance worker for your own knowledge system. Obsidian gives you the tools and the freedom to build anything. The price of that freedom is that you also do all the upkeep. Forever.

As one user on the Obsidian subreddit observed: "Sometimes it feels like people use tools like Obsidian for the sake of using the tool itself." The optimization spiral is real. You spend more time tweaking your vault than using the knowledge inside it.

MCP changes the equation (mostly)

The Model Context Protocol shifted things. Now Claude can read and write directly to your Obsidian vault through MCP servers. There are 364 MCP server repos on GitHub. The top three (mcp-obsidian, obsidian-claude-code-mcp, and obsidian-mcp-plugin) each take a different approach to connecting Claude with your notes.

Eleanor Konik, who has 12 million words in her vault, tested MCP Tools and said: "In ten minutes or less, I really did cycle through all of my old daily notes and reformat their information into skimmable log files... MCP Tools is the first time I've actually gotten an AI plugin to do what I wanted it to."

That quote matters. Someone with 12 million words of notes got an AI tool to actually work on them. The Obsidian + MCP combination is legitimate.

But the community itself is moving in a telling direction. The most popular AI plugin, Smart Connections, quietly switched from MIT to a proprietary license in January 2026 and moved features behind a $99/year paywall. The maintainer banned the user who raised the licensing issue. The community response was instructive. As one commenter noted: "It is so much more powerful and helpful to use Claude Code, Cursor, Gemini CLI or what have you, directly on your vault."

The trend is clear. In-app AI plugins are losing ground to external AI access via MCP. Obsidian users are realizing that the best way to make their vault intelligent is to point a capable AI at it from outside.

This tracks with what Tiago Forte has been saying about the shift from static knowledge bases to AI-native systems. Even the "Building a Second Brain" methodology is pivoting to AI-first in 2026. The question is no longer whether AI should touch your notes. It is what happens after it does.

Access versus understanding

This is where the distinction gets important.

When you connect Claude to an Obsidian vault via MCP, Claude gets access to your files. It can read them, search them, create new ones. That is useful. But Claude treats your vault the way it treats any collection of markdown files. It has no schema for what matters. It does not know which notes are stale, which decisions are still active, or which contacts you spoke with last week.

Your vault is flat files with whatever structure you imposed on them. Claude can parse the text, but it cannot assess the health of your knowledge base, detect what is decaying, or tell you that you have been neglecting an entire domain of your work for six weeks.

A PhD student whose project got 700 GitHub stars in 10 hours said something that stuck with me: "I didn't need Claude to remember my codebase, I needed it to tell me I've been eating like garbage for two weeks."

That is the gap. Access to files is a solved problem. Understanding of the person behind those files is not.

What I built instead

Second Brain 2.0 starts with a questionnaire that detects your archetype (founder, consultant, product manager, developer). Based on your answers, it generates a personalized GitHub repo with a CLAUDE.md file, MCP tools, and skills tuned to how you actually work.

The data layer is SQL (PocketBase locally or Supabase in the cloud), not flat markdown. This means structured records with timestamps, categories, confidence scores, and relationships between entries. When Claude reads your knowledge base, it reads structured data with metadata, not a folder of .md files with inconsistent formatting.

The system includes health scoring from 0 to 100, knowledge decay detection, and a weekly pulse that tells you what is getting stale. It migrates data from ChatGPT, Claude conversations, Google Drive, and Notion. 85 setups have been deployed so far.

Pricing is $237 one-time for the base tier, with MemoryOS (the ongoing intelligence layer) at $99 per year.

Honest trade-offs

I want to be direct about where Obsidian wins.

Flexibility and control. Obsidian lets you build exactly the system you want. Local markdown files, 2,700 plugins, custom views and workflows that no opinionated product will match. You own everything. If you enjoy building and maintaining systems, Obsidian is probably the right choice. Genuinely.

Cost. Obsidian is free. Sync costs $48 to $96 per year. Second Brain 2.0 costs more, period.

Where Second Brain 2.0 wins is in the maintenance question. The system monitors its own health. It detects knowledge decay without you asking. It structures data so Claude can reason about patterns across your entire knowledge base, not just search for keywords in files. You do not need to be the architect, organizer, and janitor of your own knowledge system.

Who should use what

If you already have a thriving Obsidian vault and enjoy maintaining it, keep using Obsidian. Point Claude at it via MCP. You will get real value from that combination, especially as the MCP servers mature.

If your vault became a junk drawer (or you never built one because the setup cost felt too high), the problem is not your discipline. The problem is that flat-file knowledge systems require ongoing human curation that most working professionals cannot sustain. A structured, self-maintaining system solves a different problem than a flexible workspace.

And if you are the kind of person who wants Claude to understand you, not just read your files, the architectural difference matters. SQL-backed structured data with health scoring, decay detection, and automated maintenance is a fundamentally different proposition than markdown files in a folder, however well-organized.

The PhD student was right. Remembering files is table stakes. The harder, more valuable problem is building a system that knows what you need before you ask.

Within a year, every serious knowledge system will be SQL-backed with validation layers. Flat files will be what filing cabinets are now: technically functional, practically abandoned.

What does your vault look like after six months of real use? I am genuinely curious.