gbrain vs obsidian-mcp-tools: which Markdown brain should your company run?

The short answer: If your team is non-technical — or mostly non-technical — and you already use (or are willing to use) Obsidian, pick obsidian-mcp-tools. It installs from an app store in 5 minutes and needs no terminal. If you have a technical lead, you want a shared brain across the whole company (including people, companies, investors), and you're ready for a 45-minute Postgres install, pick gbrain. Both keep your memory as plain Markdown files you can read, version, and edit without an agent involved.

---

At a glance

| | obsidian-mcp-tools | gbrain | |---|---|---| | Maintainer | jacksteamdev (community) | Garry Tan (YC) | | License | MIT | MIT | | Install time | ~5 minutes | ~45 minutes | | Terminal required | No | Yes | | Database | Obsidian vault (files) | Postgres + pgvector | | Best single-person use case | Founder's personal brain | Founder's personal brain + company-wide | | Best team size | 1–10 | 10–200+ | | Search quality | Keyword + file-level | Hybrid keyword + embedding | | Production brain example | Unknown (many small orgs) | Garry Tan: 17k pages, 4k people, 723 companies | | Install guide | /memory/tools/obsidian-mcp-tools | /memory/tools/gbrain |

The philosophical overlap

Both tools answer the same question: can we give our agent a memory that's stored as Markdown files we could still read without the agent? That matters for CEOs — you don't want your company's brain trapped inside a vendor database you can't inspect, export, or git-version. Plain Markdown files in a folder (Obsidian) or in a Git repo (gbrain) are forever-legible.

Where they diverge is scale, technical surface, and search quality.

When to pick obsidian-mcp-tools

Pick this if:

• You're the only operator of the brain, or your team is ≤10 people.
• You already keep notes in Obsidian — or you're willing to migrate. (Obsidian is free, beautiful, and has a ~5 minute learning curve.)
• You have no terminal skills on the team, or you have them but don't want to maintain infrastructure.
• You want the highest-velocity start. From download to "Claude can read my SOPs" is 5 minutes.
• Your search needs are "find notes that mention X" not "find patterns across my entire relationship history."

This is the path for about 80% of founder-led companies under 50 people. Start here. You can migrate later.

When to pick gbrain

Pick this if:

• You have someone on the team who's comfortable with git, docker, and npm.
• You track a lot of people and companies — investors, customers, partners, candidates — and you want your agent to reason across all of them ("who have we met with in Q2 that works on compliance?").
• You want a single shared brain that every employee's agent reads from (not per-person vaults).
• You want Git history on your organizational memory — who added what, when, why.
• You want the actual reference implementation a sitting CEO runs his company on.

gbrain is the serious version. It's not hard — but it is not 5 minutes.

The 80/20

If the install-time question feels overwhelming, this is the heuristic:

If every person in your company has ever written a Markdown file, start with obsidian-mcp-tools. If you need one brain across the whole company and someone technical can stand it up, pick gbrain.

What if we have both?

Not uncommon. A pattern we've seen:

• Founder runs gbrain on their laptop for deep personal work — investors, customers, hiring pipeline.
• Each team (CX, ops, eng) uses obsidian-mcp-tools on their own shared vaults for role-specific docs and SOPs.

They don't collide. They serve different questions.

What's the same

• Both expose MCP to any client — Claude Code, Cursor, Claude Desktop, Gemini CLI.
• Both keep your data local by default. No cloud, no vendor lock-in.
• Both pair cleanly with Cognition CLO on top — CLO doesn't care which Markdown brain is underneath; it tracks which concepts your team is about to forget.

What's different in daily use

Finding things: obsidian-mcp-tools gives your agent file-level retrieval — it reads whole notes and scans them. That works great up to maybe 500 notes. Past that, an agent with a bigger vault starts struggling with "needle in a haystack" queries. gbrain's pgvector hybrid search handles 10k+ pages without degrading.

Cross-entity reasoning: gbrain was designed around people and companies as first-class entities. If you ask "show me everyone we've talked to this month who's interested in fundraising help," gbrain shines. obsidian-mcp-tools can do it, but it leans harder on your file organization being good.

Writing back: Both let your agent add and edit notes. In obsidian-mcp-tools, new notes appear instantly in your open vault. In gbrain, new notes need a re-index step (a one-line command) before they're searchable by embedding. You can automate this with a git hook.

The CEO lens

Here's the honest advice: if you're a founder and you've never run docker compose in your life, don't start with gbrain. Even if it's "the better one." obsidian-mcp-tools will give you 80% of the value in 5 minutes, and you will actually use it. gbrain sitting unconfigured for three weeks because you never got around to it is worth zero.

Pick the one you'll actually ship this afternoon.

FAQ

Can I migrate from obsidian-mcp-tools to gbrain later?

Yes — your Obsidian vault is already Markdown files. Copy them into gbrain/content/, run npm run index, and they're live. The migration cost is just the infra setup.

Does gbrain require OpenAI embeddings?

It uses an embedding model for the vector search side. OpenAI is the default, but you can swap it for a local model or a cheaper provider. See the repo for config.

How does obsidian-mcp-tools handle conflicts if two people edit the same note?

It doesn't — Obsidian's sync (via iCloud, Dropbox, or Obsidian Sync) handles conflicts. For real multi-user writing, consider obsidian-mcp-plugin (the HTTP variant) or move to gbrain.

Is gbrain HIPAA-compliant?

The repo itself isn't configured for compliance out of the box. You can self-host on infrastructure that is compliant — Postgres + pgvector is standard, plenty of HIPAA-aware Postgres providers exist. Don't ship PHI to an LLM without a BAA.

Does the registry recommend one over the other based on my company?

Yes — the /stack recommender picks between them (and other memory options) based on your company description. Try it; takes 2 minutes.

---

Full install guides: obsidian-mcp-tools · gbrain.