io.github.Ranork/remnus (MCP) Server

An MCP-native, open-source workspace alternative to Notion. It supports reading and writing of pages, databases, and kanban boards. The server is intended to let both users and AI agents access workspace content through MCP.

🛠️ Key Features

Read & write pages
Read & write databases
Read & write kanban boards
Labeled as “MCP-native” and “open-source”

🚀 Use Cases

Workspace management using pages, databases, and kanban boards
AI agents (e.g., Claude, Cursor) reading and writing via MCP
Human-and-agent collaboration on the same workspace content

⚡ Developer Benefits

Integration path for AI agents using MCP to interact with workspace structures
Consistent capabilities across pages, databases, and kanban boards

⚠️ Limitations

Description provided does not specify authentication, deployment options, or configuration details

📚 Topics

ai
mcp
workspace

Remnus

Open-source MCP-native workspace for humans and AI agents.

Kanban boards, databases, and pages that Claude, Cursor, and any AI agent can read and write via MCP — alongside you.

smithery badge

What is Remnus?

Remnus is the Human-Agent Collaborative Workspace, built around the Model Context Protocol (MCP). Every page, database, and kanban board in your workspace is accessible to AI agents via a first-class MCP server — connect editors with one-click OAuth 2.1 (PKCE), or use a scoped bearer token for headless / CI agents.

Unlike Notion's MCP integration, Remnus is designed for headless, CI/CD, and coding agent workflows from day one.

OKF-native context, not a workspace dump

Remnus uses Open Knowledge Format (OKF) ideas as a portable knowledge contract. Pages can carry descriptions, types, tags, sources, lifecycle state, freshness dates, and revision-bound review signals. OKF v0.2 import/export is the interchange layer; the editable Remnus workspace remains canonical.

The prepare_context MCP tool turns a concrete task into Context Pack v2: a token-budgeted set of concepts ranked by BM25 relevance, native metadata, page links, freshness, and trust. Smart mode advertises this context-first flow automatically for meaningful work. Strict mode additionally requires the same agent's short-lived contextRunId before Remnus MCP mutations.

This does not let an MCP server control local files, shell commands, or Git. Repository-level enforcement uses the supplied AGENTS.md and CLAUDE.md snippets. Read the public Context-First MCP guide and the implementation story, How Remnus Uses OKF to Give AI Agents Better Context.

Features

Pages — Markdown editor with slash commands, nested sub-pages, and icons
Databases — Customizable columns, Table / Kanban / Calendar views, filters, sorts
MCP Server — 24 tools + 6 resources + 7 prompts over stateless Streamable HTTP
OKF-native context — Portable knowledge metadata, exact-revision reviews, Context Pack v2, and Smart/Strict MCP policies
Agent auth — One-click OAuth 2.1 + PKCE (RFC 7591 dynamic registration) for editors, or scoped read/write personal access tokens for headless agents
Multi-workspace — Invite members, role-based access (owner / member / viewer)
Desktop app — Tauri v2 shell for Windows, macOS, Linux
Mobile — Capacitor v8 for iOS and Android (loads remnus.com)
i18n — English, Türkçe, हिन्दी, Español, Français, Deutsch, 中文, Русский

Quick Start — Self-host

Local Development

bash

git clone https://github.com/Ranork/remnus-app.git
cd remnus-app
cp .env.example .env          # fill in AUTH_SECRET + OAuth credentials
npm install
npm run db:migrate
npm run dev

Open http://localhost:3000. The first user to sign up is auto-promoted to admin.

Docker Compose (5-minute setup)

If you prefer to run Remnus using Docker:

Clone the repository and navigate into it:

bash

git clone https://github.com/Ranork/remnus-app.git
cd remnus-app

Copy the environment template and fill in the required variables (especially AUTH_SECRET and OAuth credentials):
bash
```
cp .env.example .env
```
Start the application:
bash
```
docker compose up -d
```
Access Remnus at http://localhost:3000. The SQLite database will be persisted automatically using a Docker volume.

Deploy

Deploy to Railway

Add MCP to your editor

After signing in, open the AI Agents panel from the sidebar and click Connect editor. It walks you through every supported editor (Cursor, VS Code, Claude Code, Codex, Windsurf, and more).

Install from a registry

Remnus is published on the official MCP Registry (io.github.Ranork/remnus) and on Smithery, so MCP-aware clients and directories can discover it automatically. Smithery offers one-click install across many clients (Claude, Cursor, VS Code, Codex, Windsurf, and more) — open the Smithery page and click Add to toolbox, or use the Smithery CLI to install straight into a client's config:

bash

npx -y smithery@latest mcp add ranorkk/remnus --client claude

Swap --client for cursor, vscode, codex, windsurf, etc. Either way, the first connection runs the same OAuth 2.1 sign-in — no token to paste.

Recommended — OAuth (token-less): point your editor at the MCP URL and approve the consent screen on first connect. Your editor runs the OAuth 2.1 flow automatically — nothing to copy.

json

{
  "mcpServers": {
    "remnus": {
      "type": "http",
      "url": "https://your-instance.com/api/mcp"
    }
  }
}

Headless / CI — personal access token: create a scoped token from the AI Agents panel and send it as a bearer header.

json

{
  "mcpServers": {
    "remnus": {
      "type": "http",
      "url": "https://your-instance.com/api/mcp",
      "headers": {
        "Authorization": "Bearer <your-token>"
      }
    }
  }
}

MCP Tools

Tool	Scope	Description
`prepare_context`	read	Build Context Pack v2 (BM25 + OKF metadata + graph) and return a short-lived `contextRunId`
`search_workspace`	read	Full-text search across pages and databases
`list_workspace`	read	List sidebar items with pagination
`get_page`	read	Get a page or database row by ID
`get_pages`	read	Get multiple pages/rows by ID in one call (batch `get_page`)
`get_database_schema`	read	Get column schema of a database
`query_database`	read	Query rows with filters and pagination
`list_members`	read	List workspace members with roles
`query_audit_log`	read	Filtered agent activity log
`get_changes_since`	read	Compact, chronological feed of what changed since a timestamp or cursor
`get_related_pages`	read	A page's parent, children, outgoing links, backlinks, and database siblings
`create_page`	write	Create a standalone page or database row
`update_page`	write	Update title, content, or properties
`bulk_update_pages`	write	Update multiple rows in one call
`delete_page`	write	Delete a page (requires `confirm: true`)
`bulk_delete_pages`	write	Delete up to 100 pages/rows/databases in one call (requires `confirm: true`)
`move_item`	write	Move item to a new parent
`bulk_move_items`	write	Move up to 100 items in one call, reparenting or moving rows to a different database
`create_database`	write	Create a database with custom schema
`update_database_schema`	write	Add or remove columns
`create_database_view`	write	Add a table/kanban/calendar view
`update_database_view`	write	Rename a view or patch its config
`delete_database_view`	write	Delete a saved view (requires `confirm: true`)
`add_comment`	write	Add a comment to a page or row (agent comments are append-only)

Tech Stack

Framework: Next.js 16.2.6 (App Router)
Database: SQLite via Drizzle ORM + @libsql/client (Turso-compatible)
Auth: Auth.js v5 — Google & GitHub OAuth
Styling: Tailwind CSS + Lucide icons
Desktop: Tauri v2 (Rust)
Mobile: Capacitor v8

Contributing

See CONTRIBUTING.md. All contributions are welcome — bug fixes, new MCP tools, translations, and docs.

License

AGPL-3.0 — free to self-host and modify. SaaS forks must open-source their changes.

io.github.Ranork/remnus