GitHub
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
作者による説明(英語の原文)
Capture dev context on interrupt; replay it as a 30-second resume card. Git stash for your head.
この3つを1つのスコアにまとめることはしません。意見、測定、人気はそれぞれ別のものだからです。
このサーバーのやさしい解説はまだ書いていません。以下はレジストリから読み取った事実です。
使っているAIアプリを選んで、設定をコピーしてください。
必要なもの
どのアプリを使っていますか?
上でアプリを選んでください。選んだアプリを覚えておくので、次からはボタンに「…にインストール」と表示されます。
Claude Desktop → Settings → Developer → Edit Config を開き、claude_desktop_config.json にこれを貼り付けます
{
"mcpServers": {
"br8n": {
"command": "uvx",
"args": [
"br8n",
"--from",
"br8n",
"br8n-mcp"
]
}
}
}ターミナルでこれを実行します
claude mcp add br8n -- uvx br8n --from br8n br8n-mcpプロジェクトの .cursor/mcp.json、または ~/.cursor/mcp.json に貼り付けます
{
"mcpServers": {
"br8n": {
"command": "uvx",
"args": [
"br8n",
"--from",
"br8n",
"br8n-mcp"
]
}
}
}ブラウザがアプリを開いてよいか確認し、そのあとアプリが設定を表示して確認を求めます。キーの値は空欄のままなので、自分で入力してください。
プロジェクトの .vscode/mcp.json に貼り付けます
{
"servers": {
"br8n": {
"type": "stdio",
"command": "uvx",
"args": [
"br8n",
"--from",
"br8n",
"br8n-mcp"
]
}
}
}ブラウザがアプリを開いてよいか確認し、そのあとアプリが設定を表示して確認を求めます。キーの値は空欄のままなので、自分で入力してください。
これはパソコンで動くサーバーですが、ChatGPTがつなげるのはインターネット上のサーバーだけです。ほかのアプリを選んでください。
選んだアプリは、この端末のCookieに保存されます。アカウントは不要です。
この設定は、作者のレジストリ登録情報から作成しています。READMEと違う場合は、READMEに従ってください。
パソコンで動くサーバーです。インストールすると、ツール一覧がAIアプリに表示されます。
誰でも見られる情報をチェックしており、コードは読んでいません。当てはまらない項目は数に入れていません。 · 2026年11月7日にチェック
レジストリ名は、リポジトリを所有するGitHubアカウント anthonysuherli で確認済みです
このリポジトリはまだGitHubから読み取っていません
このリポジトリはまだGitHubから読み取っていません
これはパソコンで動くため、インターネット上のエンドポイントはありません
パソコンで動き、代わりにローカルの設定を使います
パソコンで動くサーバーです。ツール一覧を読み取れるのは、インターネット上のエンドポイントがある場合だけです
レジストリでこのカテゴリにある有効なすべてのサーバーと比較しています。スターは、公開者がリポジトリの所有者である場合だけ数えます。
当てはまる基本チェック1項目のうち1項目に合格しました。コードは読んでいません。キーやトークンを追加する前に、サーバーがどんなアクセスを求めているかを確認し、必要な権限だけを持つキーを使ってください。
レジストリには必須のキーは載っていません。ただし、最初に使うときにサインインを求めるサーバーもあります。
あなたのパソコンで動きます。必要なときにAIアプリが起動し、データは作者のサーバーを通りません(サーバー自体がオンラインサービスを呼び出す場合を除きます)。
GitHubリポジトリの最終更新は–、最新バージョン(1.1.2)がレジストリに届いたのは3 か月前です。
Claude Desktop、Claude Code、Cursor、VS CodeなどのMCP対応アプリで使えます。パソコンにPython (uv)も必要です。
同じカテゴリには:GitHub, Desktop Commander, Cloudflare MCP
同じカテゴリで、ほかの作者が作ったサーバー。スターの多い順。
カードの見方:★ GitHubスターi3つの柱:レビュー · チェック · 利用状況i更新=最新バージョンをレジストリに公開した日i
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
MCP server for terminal commands, file operations, and process management
Cloudflare MCP servers
作者のリポジトリから書かれたとおりにコピーしたものです。PopNx MCPは内容を確認・推奨していません。HTML、画像、スクリプトは取り除いています。リンクはこのサイトの外へ移動します。
Site · install · how it works · pip install br8n
You'll forget what this branch was for. br8n won't. It saves what you were thinking the moment you get pulled away — branch, open files, the diff, and the one-line *why* — and hands it back as a 30-second resume card when you return. Your code is already saved; this is the part that isn't. Research puts the cost of refocusing after an interruption at about 23 minutes (Gloria Mark, UC Irvine), and longer for complex code. *(Snapshots are taken two ways today: automatically at commit boundaries via an installed post-commit hook, and on demand with /br8n:capture. There is no continuous watcher.)*
Most tools capture *state* (files, layout, git history). br8n captures *intent* — the one-line hypothesis in your head: *"JWT validation is caching stale tokens."* That's the wedge that matters.
Beyond your current device, br8n is a portable knowledge engine: your captured insights live in a searchable journal accessible from Claude Code, the iOS companion, or any tool that speaks HTTP. Sync, search, and share across machines (paid tier, future).
Before a meeting, a branch switch, or end of day, run /br8n:capture. br8n snapshots your workspace in one second and records *"What were you working on?"* — the one-line hypothesis is the load-bearing field.
Before: Capture: After:
┌─────────────────┐ /br8n:capture ┌──────────────────┐ │ Fixing bug in │ ───► br8n asks: ───► │ Finding saved: │ │ auth flow │ "What were │ │ • git diff │ │ files: [3] │ you doing?" │ │ • open files │ │ branch: fix-#42 │ Fixing auth │ │ • cursor pos │ └─────────────────┘ bug │ │ • hypothesis │ └──────────────────┘ (stored in KB)
Your captured snapshots live in a searchable journal. One hypothesis per snapshot— the thing you'd write on a post-it.
Open br8n (or focus your editor). The resume card appears instantly with:
╔════════════════════════════════════╗
║ br8n — Where were you? ║ ╠════════════════════════════════════╣ ║ 📌 Fixing auth bug in login flow ║ ║ ║ ║ Recent snapshots: ║ ║ • 5 min ago: auth middleware ║ ║ • 12 min ago: jwt validation ║ ║ • 45 min ago: session storage ║ ║ ║ ║ Coverage: ████░ (rich) ║ ╚════════════════════════════════════╝
No digging through git logs. No "where was I again?" Back to work in 30 seconds.
If coverage is gap (you've been away a while, or switched branches), one click runs a web-research pipeline to pull in fresh context: changed docs, new issues, updated deps—and folds it back into your session knowledge base.
Resume card says "coverage: gap"
│ ▼ ┌─────────────────┐ │ [Explore Now] │ ─► web search (changed deps, docs) └─────────────────┘ ─► fetch + parse relevant sources │ ─► extract + embed findings ▼ Coverage updates to "rich" + new context appears in the card
Perfect for returning after a weekend or after your teammate merged a big change.
br8n is a self-contained fork of Delapan, repurposing its primitives (Findings, pgvector search, tap/preamble) from chat to automatic capture.
Slash commands from inside any Claude Code session:
/br8n:pickup → Show the current repo/branch resume card
/br8n:capture → Save a snapshot right now /br8n:search → Ask a question, grounded in your session history /br8n:explore → Force the gap-fill pipeline
Example: You're in a Claude Code session debugging auth. Type /br8n:search "how did I set up JWT validation?" and get an answer from your captured snapshots.
A native SwiftUI app — the read spine. Sign in with Apple, browse your cross-repo activity, and read resume cards from your phone. Consumes the same /v1/projects + /v1/resume + /v1/activity API, authenticated per-user (see below).
Your captured snapshots form a searchable knowledge journal. The engine runs in two tiers from the same code — the difference is where your data lives.
| Tier | Free / local | Paid / cloud |
|---|---|---|
| Storage | On-device SQLite | Hosted Supabase (pgvector) |
| Access | Loopback only (localhost:8002) | Anywhere (with API key) |
| Sign-in | None | GoTrue |
| Data | ~/.br8n/brain.db | Encrypted, RLS protected |
| Select | BR8N_BACKEND=local | BR8N_BACKEND=cloud + creds |
Access your knowledge anywhere:
Claude Code (local/cloud) iOS companion (cloud) Browser (cloud, future)
│ │ │ └──────────────┬───────────────┴───────────────────────┘ │ br8n API │ ┌───────┴────────┐ │ │ SQLite Supabase (local db) (cloud db)
Free tier: single device, no sync. Paid tier: access from Claude Code, the iOS app, or any tool that speaks HTTP. Team sharing and cross-repo search are designed (future).
The paid value props — cross-machine sync, cross-repo search, managed keys, team sharing — are not yet shipped.
14:32 — Debugging auth middleware
Open file: middleware.py, line 45 Hypothesis: "JWT validation is caching stale tokens"
14:35 — [Meeting call] /br8n:capture → br8n saves the snapshot
15:47 — [Back from meeting] /br8n:pickup → resume card appears: "🔸 JWT validation is caching stale tokens" Recent context shown. No "where was I?" moment.
You're on fix/session-timeout, about to switch to main
/br8n:capture → snapshot saved against this branch
Hours later, switch back: git checkout fix/session-timeout /br8n:pickup → resumes from that branch → shows the last hypothesis + snapshots
You're in a Claude Code session, ask a question:
/br8n:search "how did I set up the JWT secret?"
Claude Code queries your captured session history and answers from your own notes/decisions.
Install the Claude Code plugin — no venv, no path editing. The plugin's MCP server bootstraps its own environment on first run.
/plugin marketplace add anthonysuherli/br8n
/plugin install br8n@br8n
Reload the session, then use /br8n:pickup, /br8n:capture, etc. Data lives in ~/.br8n/brain.db on the free/local tier.
What works without any key: capture and resume. Snapshots are stored without embeddings, so the resume card, the hypothesis and the snapshot trail all work with no account and no key at all.
What needs a key: semantic search (/br8n:search) needs an embedding key — AI_GATEWAY_API_KEY or OPENAI_API_KEY. The explore / gap-fill pipeline needs that *and* TAVILY_API_KEY for web search.
Keyless semantic search: on the local tier, pip install 'br8n[local-embeddings]' gives you semantic search with no API key at all — an on-device ONNX model (bge-small-en-v1.5, ~130MB, no torch, downloaded once on first use). Use /br8n:embeddings to check which provider is active or switch between them. A keyless python -m br8n.vault.reindex rebuilds real vectors, not just text.
Vectors from different models aren't comparable, so br8n keeps one active embedding space and re-embeds when you change providers. Because a provider can also change *by accident* — a shell without your API key resolves to the local model — br8n will not silently throw your vectors away: a change it inferred from the environment is offered, not applied. /br8n:embeddings (and --check) report the pending switch, and it happens when you confirm. While an offer is pending, capture and search behave exactly as they do with no key at all: nothing breaks, nothing is lost, and the backlog re-embeds itself once you confirm. A switch you ask for explicitly applies immediately.
Sanity-check a local install with python -m br8n.api.main --check — it reports your Python version, which backend tier is configured, whether sqlite-vec loads, whether the DB path is writable, and whether the embedding and explore keys are present. Anything missing is named explicitly.
Free/local (SQLite, single device):
BR8N_BACKEND=local python -m br8n.api.main # listens 127.0.0.1:8002
Paid/cloud (Supabase, accessible anywhere):
BR8N_BACKEND=cloud uvicorn br8n.api.main:app --reload --port 8002
Working on br8n itself, rather than using it:
git clone https://github.com/anthonysuherli/br8n
cd br8n/backend python3.11 -m venv .venv .venv/bin/pip install -e ".[dev]" cp .env.example .env
pip install br8n also works as a plain package install, and ships the br8n-mcp and br8n-server entry points; the plugin marketplace remains the supported path for using br8n inside Claude Code.
By default, the knowledge base is keyed by project (git repo name) and kb (git branch). Override the database path with BR8N_DB_PATH (local) or set Supabase credentials in .env (cloud).
The cloud tier is multi-user: each request carries a Supabase GoTrue JWT — obtained via Sign in with Apple through POST /v1/auth/apple (Supabase verifies the Apple token and provisions the user) and rotated via POST /v1/auth/refresh. The backend verifies the JWT against SUPABASE_JWT_SECRET and scopes every read/write (findings *and* the activity graph) to the caller's own org via row-level security. BR8N_API_KEY remains as a service-only key for internal callers. The local tier needs no auth (loopback-only, single user).
See CLAUDE.md for:
br8n/ core engine, fork structure)/v1/capture, /v1/resume, /v1/explore, /v1/auth/apple)/br8n:pickup, etc.)/v1/auth/apple + /v1/auth/refresh冒頭だけを表示しています。続きはリポジトリで読めます