Open Brain MCP server
Commit Nov 14, 2026Provides a personal AI memory system that captures thoughts, manages contradictions, and compiles wiki pages via an MCP server, CLI, and Slack bot.
Also known as dagonet/open-brain
- Stars
- 8
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 14, 2026
- License
- MIT
What is Open Brain MCP?
Open Brain provides a personal AI memory system that captures thoughts, manages contradictions, and compiles wiki pages via an MCP server, CLI, and Slack bot.
Summary generated from the project's README
About Open Brain MCP server
Capturing thoughts from a shell, Claude Code, or Slack to create a unified personal memory is what open-brain enables. It allows developers to store decisions, insights, and meeting notes in a way that remains searchable by meaning rather than just keywords. By using an MCP server, a CLI, and a Slack bot, you can ensure that every AI tool you use accesses the same context.
Users manage their own data through a Supabase project and use OpenAI for embeddings and classification. The system automatically identifies contradictions between notes, such as conflicting technical decisions, and allows you to resolve them to maintain accuracy. It also compiles readable wiki pages that cite specific notes and track when information becomes stale.
This setup includes specialized tools like thoughts_capture for saving information, entities_search for exploring the relationship between people and projects, and a suite of task tools including task_create and task_list. A web dashboard provides a visual way to browse the entity graph and manage contradictions.
What you can do with it
- 1Capture context from multiple interfacesSave a thought via the command line, a Slack message, or directly through Claude Code.
- 2Resolve conflicting technical decisionsIdentify and fix instances where older notes disagree with newer information.
- 3Generate documented topic summariesTurn a collection of scattered notes into a structured wiki page with citations.
- 4Track action items within AI workflowsManage tasks and their status history using specialized command tools.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachabledagonet/open-brain is public and reachablegithub.com/dagonet/open-brain ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-12github.com/dagonet/open-brain ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubOpen Brain
Give every AI tool you use the same memory: capture a thought once, from the shell, Claude Code or Slack, and find it again by meaning.
Open Brain is for people who run Claude Code on real work and want a memory they own. Your data lives in your own Supabase project. Note text is sent to OpenAI for embeddings and classification (on capture, wiki compile and the contradiction audit).
What you get
Capture from anywhere
Type brain "..." in a shell, let Claude Code call thoughts_capture, or message a Slack bot. There is no folder or filename to think about. Every note is embedded, classified (decision, insight, meeting, action, reference, question or note), tagged with its people, topics and action items, and rated for how much it matters.
Search that ranks what matters
Ask "what did I decide about X last quarter?" and the right notes come back by meaning, not keyword. Results blend relevance with recency, importance and contradiction status, so a fresh decision beats a stale note. Memory can be scoped per repo, near-duplicates are flagged so you can supersede them, and archived thoughts stay out of the way.
Wiki pages with provenance
For any topic you have a few notes on, Open Brain compiles one readable page. Every paragraph cites the notes it came from, and the page tells you when it has gone stale. Refresh it from the CLI, from Claude or from the dashboard.
Your own contradictions
A scan finds pairs of notes that disagree, for example an old "we picked Postgres" next to a newer "we switched to SQLite". You decide which one is current. While a contradiction is open, the wiki leaves out both notes; once you resolve it, both come back.
An entity graph
People, projects and technologies you mention become a co-occurrence graph. Explore it with entities_search and entities_graph, widen a search along shared entities with thoughts_search_expanded, or look at it on the dashboard's /graph page.
Tasks
task_create, task_get, task_list and task_update track action items from Claude Code, with a status history. A token-protected tasks-feed gives a local tool a read-only list.
A dashboard
Browse thoughts, wiki pages, contradictions and the graph in a browser. It is private to you and deploys to Vercel.
It looks after itself
Nightly jobs audit contradictions, recompile stale wiki pages and archive old notes, and a weekly job consolidates topics, each with an LLM budget cap.
You own all of it
The data lives in your Supabase project. The public key can read nothing, the dashboard is scoped to your account, and the server key never leaves your own tools. See docs/security.md.
What it looks like
The output below is illustrative; your notes and the model's answers will differ.
From a shell:
$ brain "Met with Sarah today, she's considering consulting"
✓ Thought captured
Type: decision
People: Sarah Johnson
Topics: consulting
Action: Follow up with Sarah about consulting timeline
In Claude Code, which searches your memory on its own at the start of a session and when a person or project comes up:
> What did we decide about the build tool?
thoughts_search("build tool decision") -> 3 notes, best match first
We switched to esbuild in March because ... (your own words, with the note it came from)
On the dashboard, after you sign in:
| Route | What it shows |
|---|---|
/ |
Your thoughts, filterable by topic, person and type, with text search |
/wiki |
Compiled topic pages, each with its sources and a staleness banner |
/contradictions |
Pairs of notes that disagree, with a resolve form |
/graph |
Contradictions and entities as force-directed graphs |
Slack: invite the bot to a channel and every message people post there is captured (bot posts and edits are ignored), with a threaded reply that shows what was extracted.
Get started
You need Node.js, the Supabase CLI, a Supabase project and an OpenAI key. The short version:
supabase link --project-ref <project-ref> && supabase db push --remote # 1. the database
supabase functions deploy capture-thought # 2. the functions, each by name (six in all)
npm ci --prefix cli && npm run build --prefix cli && npm install -g --install-links=false ./cli # 3. the brain command
On Supabase CLI 2.x, use --linked instead of --remote in step 1 (see docs/install.md).
That is a teaser, not the guide. The full walk-through, with keys, secrets, the MCP server, the dashboard and optional Slack and nightly jobs, is in docs/install.md. You can also try everything on a local stack first.
Upgrading
Coming from 0.8: docs/upgrading.md. Keys first, then the new migrations, then the functions by name. Release notes for every version are in CHANGELOG.md.
Learn more
| Document | What is in it |
|---|---|
docs/install.md |
Fresh install, step by step, and how to verify it |
docs/upgrading.md |
Moving to a new release |
docs/cli.md |
The brain command |
docs/mcp.md |
The MCP server and its 22 tools |
docs/dashboard.md |
The web dashboard |
docs/edge-functions.md |
The six edge functions |
docs/configuration.md |
Every environment variable |
docs/security.md |
Who can do what, keys, and the lockdown |
docs/operations.md |
Nightly jobs, routine upkeep, checks |
docs/architecture.md |
How it fits together and the data model |
docs/troubleshooting.md |
Symptoms and fixes |
docs/background/ |
Background: a summary of the video that inspired the project (not of Open Brain itself), in English and German |
Already using claude-code-toolkit? Its templates ship with the Open Brain integration pre-wired, and OPEN_BRAIN_TOOLS_DISABLED in a repo's .mcp.json silences tool families you do not want there (docs/mcp.md).
Inspired by
- Andrej Karpathy — LLM Wiki gist (the upstream "personal wiki maintained by AI" idea — 41 k bookmarks)
- Nate B Jones — Karpathy's Wiki vs Open Brain (the bridge that adapted Karpathy's idea for Open Brain and announced the wiki + contradictions improvements)
- Nate B Jones — You Don't Need SaaS. The $0.10 System That Replaced My AI Workflow
- Nate B Jones — One Simple System Gave All My AI Tools a Memory. Here's How.
Related projects
Part of an ecosystem for AI-assisted development with Claude Code:
- claude-code-toolkit -- Template system for bootstrapping projects with Claude Code configuration, including pre-wired Open Brain integration
- mcp-dev-servers -- Six custom MCP servers (61 tools) for git, GitHub, .NET, Rust, Ollama, and template-sync integration
Contributing, security and license
See CONTRIBUTING.md, CODE_OF_CONDUCT.md and SECURITY.md. MIT licensed, see LICENSE.
Open Brain MCP questions
Where is my data stored?
Your data lives in your own Supabase project, ensuring you own your memory.
How can I capture thoughts?
You can capture thoughts using the 'brain' command in a shell, letting Claude Code call thoughts_capture, or by messaging a Slack bot.
What happens if I have conflicting notes?
The system performs a contradiction audit to find pairs of notes that disagree, allowing you to decide which one is current.
Can I see a visual representation of my data?
Yes, you can explore an entity graph of people, projects, and technologies via the dashboard's /graph page.