Power your agent
การติดตั้งจากผู้เผยแพร่ ไม่ใช่เอนด์พอยต์ของ PlugMCP
URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร
รันบน PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
เปิดจากผู้เผยแพร่
An AI assistant can discuss your plans. It cannot maintain the system where those plans live.
This closes that gap. It is a self-hosted MCP server that gives an AI app — Claude, ChatGPT, or any MCP client with remote connectors — read and write access to exactly one GitHub repository: your Markdown notes, typically an Obsidian vault.
Authenticated remote MCP server for Claude, ChatGPT, and other AI apps to safely read and write one GitHub-backed second-brain vault — single-user or team, with per-member commit attribution.
The assistant reads the current state of your notes, proposes a change, and writes it back. Every change is an ordinary Git commit — reviewable, revertible, and readable without any of the tools involved.
You run it yourself. There is no hosted service and no account: you deploy the server, point it at your own repository, and connect your AI app to it.
Background article: Beyond Chat History: Building an AI-Native Second Brain with Obsidian, GitHub, and MCP
Once connected, you talk to your assistant normally and it works on the real files:
Afterwards you open Obsidian, or GitHub, and see exactly what changed.
AI client (Claude, ChatGPT, …)
→ OAuth login against this server
→ remote MCP server
→ member's fine-grained GitHub PAT (server-side only)
→ one Markdown vault repository
→ Git history
Access and behaviour are two separate layers. This server controls what the
assistant is allowed to touch. The vault's own root-level CLAUDE.md — plain
Markdown, versioned alongside the notes — defines how it should behave:
folder conventions, frontmatter contracts, what it must ask about before doing.
The github_write_file tool points the assistant at that file explicitly.
Giving an AI app write access to your knowledge base is only reasonable if the access is bounded. These are the boundaries:
owner/repo. No tool accepts an arbitrary owner, repo, or URL.main, capped at 100 KB. No patch mode,
no branch sprawl — and a sha precondition, so a write fails rather than
clobbering a change made since the assistant last read the file.Full auth and threat model:
docs/security-notes.md.
OAuth persistence and deployment options:
docs/oauth-storage.md.
cp .env.example .env
# edit .env — each variable is documented inline
uv sync
uv run python server.py
Then validate the OAuth gate and connect it to your AI app (Claude, ChatGPT, …)
following docs/connector-setup.md.
All configuration is via environment variables (see .env.example):
MCP_PUBLIC_BASE_URL / OAUTH_ISSUER_URL — the server's own public URL.TEAM_MEMBER_IDS + a MEMBER_<ID>_* block per member — the roster. One entry
for single-user, several for a team.GITHUB_ALLOWED_OWNER / GITHUB_ALLOWED_REPO — the one repo everyone reads
and writes.MCP_SERVER_NAME (optional) — the name this instance reports (login page,
ping). Defaults to second-brain-github-mcp.OAUTH_DB_PATH (optional) - enables restart-safe SQLite storage for OAuth
clients, tokens, authorization codes, and login state. Without it, the server
uses memory and prints a startup warning.The app builds from the included Dockerfile and runs on any Docker host
(e.g. DigitalOcean App Platform, or a VPS via docker compose). Set every
variable from .env.example as an encrypted platform env var — never upload a
.env file — and set the public-URL vars to the real assigned URL. The server
reads env only at startup, so redeploy after any change (PAT rotation, new
member, URL). For a single production container, mount a host directory and set
OAUTH_DB_PATH=/data/oauth.sqlite3. Full steps are in the connector-setup and
OAuth-storage docs. A complete generic starting point is included as
docker-compose.example.yml.
Use /health for uptime monitors. It is intentionally unauthenticated and
returns HTTP 200 when the server process is running:
https://<your-domain>/health
uv run pytest
uv.FastMCP, Streamable HTTP transport).Built by Manuel Blinkert.
Authenticated remote MCP server for Claude, ChatGPT, and other AI apps to safely read and write one GitHub-backed second-brain vault — single-user or team, with per-member commit attribution.
เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
Second Brain Github ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-10-28 Authenticated remote MCP server for Claude, ChatGPT, and other AI apps to safely read and write one GitHub-backed second-brain vault — single-user or team, with per-member commit attribution.
