MCP.so
ログイン

Open Brain

@Bobby-cell-commits

リポジトリ

Open Brain について

Self-hostable MCP memory server for AI assistants — 16 tools, hybrid search, knowledge graph, auto-dedup, staleness pruning, automated ingestion pipelines. Built on Supabase + pgvector.

基本情報

カテゴリ

メモリとナレッジ

ライセンス

MIT

ランタイム

node

トランスポート

stdio

公開者

Bobby-cell-commits

投稿者

Bobby-cell-commits

設定

以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。

{
  "mcpServers": {
    "open-brain": {
      "type": "http",
      "url": "https://<supabase-project-ref>.supabase.co/functions/v1/open-brain-mcp/<api-key>/mcp"
    }
  }
}

ツール

16

Semantic search with optional graph expansion (1-hop traversal)

Browse thoughts filtered by type, topic, person, theme, quality, time

Aggregate statistics: counts, type/theme breakdown, top topics/people

Save a new thought from any AI client (with auto-embedding)

Graph traversal from a thought (typed links: extends, contradicts, etc.)

Browse extracted entities (people, tools, projects, orgs) by frequency

AI-generated summary of themes, open loops, and next steps

Graph analysis: hubs, density, sources, co-occurrence, themes

Duplicate candidates with similarity zone histogram

Recompute all salience scores

Rewrite content (re-embeds, re-extracts metadata)

Permanent delete (cascades connections)

Resurface forgotten high-quality thoughts

Pipeline monitoring: health status, run history, merge audit

Review and act on stale thought candidates

Instructions for importing memories from other platforms

概要

What is Open Brain?

Open Brain is a self-hostable, graph-structured memory server for MCP clients (Claude, ChatGPT, and any assistant that speaks the Model Context Protocol). Thoughts flow in from Telegram, pipelines, or direct capture and are stored in a Newman-IDF weighted entity graph with PostgreSQL and pgvector. An automated Dream cycle deduplicates near-duplicates, tracks theme drift, synthesizes insights, and archives stale content. It benchmarks at 37.2% on LongMemEval.

How to use Open Brain?

Deploy Open Brain on Supabase (hosted) or with Docker Compose (self-hosted). After deployment, connect your AI client by providing the MCP endpoint URL and the generated MCP access key. Use the claude mcp add command (Claude Code) or configure the connector in Claude Desktop or ChatGPT Settings to start capturing and searching thoughts.

Key features of Open Brain

  • Graph-structured memory with typed entity relationships
  • Automated Dream cycle for dedup, insights, and archiving
  • 17 MCP tools for search, capture, analysis, and review
  • Semantic search using vector embeddings (pgvector)
  • Weekly review synthesis of themes and open loops
  • Self-hosted – you own your data

Use cases of Open Brain

  • Capture thoughts on-the-go via Telegram bot with automatic metadata extraction
  • Perform semantic search across memories using natural language queries
  • Get AI-generated weekly reviews that synthesize themes and action items
  • Analyze knowledge graph for hubs, co-occurrence, and connection patterns
  • Import memories from other platforms using the migration guide

FAQ from Open Brain

What makes Open Brain different from a flat document store?

Open Brain uses a Newman-IDF weighted entity graph with automatic entity extraction, typed relationships (extends, contradicts, is-evidence-for, supersedes), and a Dream cycle that deduplicates, tracks theme drift, and synthesizes insights – not just a flat document store.

What deployment options are available?

You can deploy Open Brain on Supabase (hosted, with scheduling via GitHub Actions) or with Docker Compose (self-hosted, with a built-in cron container). Both use PostgreSQL with pgvector.

What dependencies are required?

You need a Supabase account (or Docker for self-hosting) and an OpenRouter API key for embeddings and metadata extraction. A Telegram bot is optional but recommended for mobile capture.

How is authentication handled?

MCP endpoint access is secured with a cryptographic personal access key. Clients send this key in the x-brain-key HTTP header when connecting to the MCP server.

Where does my data live?

All data is stored in your own Postgres database with pgvector for vector similarity search. On Supabase it is managed hosting; with Docker it runs on your own machine or server. You retain full ownership.

コメント

「メモリとナレッジ」の他のコンテンツ

Notion MCP Server

awkoy

Notion MCP server for Claude, Cursor, ChatGPT & Claude Desktop. Connect AI agents to Notion via Model Context Protocol — pages, databases, blocks, comments, files.

159

Memory Bank MCP Server

alioshr

A Model Context Protocol (MCP) server implementation for remote memory bank management, inspired by Cline Memory Bank.

913

Solomd

zhitongblog

A markdown editor — and the bridge to your LLM. Local-first, MIT, ~15 MB. Bundled MCP server lets Claude Code / Codex / Cursor drive your vault directly. 14 AI providers BYOK.

425

Jupyter Notebook MCP Server (for Cursor)

jbeno

Model Context Protocol (MCP) server designed to allow AI agents within Cursor to interact with Jupyter Notebook (.ipynb) files

157

Obsidian MCP Server

cyanheads

Read, write, search, and surgically edit Obsidian vault notes, tags, and frontmatter via MCP. STDIO or Streamable HTTP.

618