Agentic CMS
# Agentic CMS
> Open-source MCP server that turns any CMS backend into an AI-agent-ready content management system.
[](LICENSE)
## What is this?
Agentic CMS is a [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) server that gives AI agents full access to your content management workflows — create, read, update, publish, and track content through a standardized protocol.
Instead of building AI features *into* your CMS, Agentic CMS wraps *around* it. Your CMS stays as-is. The agents get a clean interface to work with.
```
┌─────────────────┐ MCP Protocol ┌──────────────┐
│ AI Agents │ ◄──────────────────► │ Agentic CMS │
│ │ stdio / SSE │ MCP Server │
│ · Claude │ │ │
│ · OpenClaw │ │ Adapters: │
│ · Cursor │ │ · Supabase │
│ · Any MCP client│ │ · Payload │
└─────────────────┘ │ · Strapi │
│ · (yours) │
└───────┬───────┘
│
┌───────▼───────┐
│ Your CMS DB │
└───────────────┘
```
## Why?
- **Your CMS, your data** — Self-hosted, no vendor lock-in
- **Adapter pattern** — Supabase today, Payload/Strapi/anything tomorrow
- **MCP standard** — Works with Claude Desktop, OpenClaw, Cursor, and any MCP-compatible client
- **Safety first** — Publishing requires human approval by default. Agents create drafts, humans publish.
- **Open source** — MIT licensed. Use it, fork it, extend it.
## Features
### Tools (MCP)
| Tool | Description |
|------|-------------|
| `list_contents` | List content with filters (status, category, tags) |
| `get_content` | Get a single content item by slug or ID |
| `create_content` | Create new content (always starts as `draft`) |
| `update_content` | Update content fields (title, body, tags, etc.) |
| `list_ideas` | List content ideas |
| `promote_idea` | Promote an idea to a draft content item |
| `create_publication` | Record a publication event (channel, URL, metrics) |
| `get_metrics` | Get performance metrics for content |
### Safety
- `create_content` always sets status to `draft` — agents cannot publish directly
- `update_content` blocks status changes to `published` — human approval required
- All operations are logged and auditable
## Quick Start
### 1. Install
```bash
npm install @brxce/agentic-cms
```
### 2. Configure
Create `.env`:
```env
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key
```
### 3. Run
```bash
npx agentic-cms
```
### 4. Connect to Claude Desktop
Add to `claude_desktop_config.json`:
```json
{
"mcpServers": {
"agentic-cms": {
"command": "npx",
"args": ["@brxce/agentic-cms"],
"env": {
"SUPABASE_URL": "https://your-project.supabase.co",
"SUPABASE_SERVICE_ROLE_KEY": "your-key"
}
}
}
}
```
### 5. Connect to OpenClaw
```yaml
# ~/.openclaw/config.yaml
mcp:
servers:
agentic-cms:
command: npx
args: ["@brxce/agentic-cms"]
env:
SUPABASE_URL: https://your-project.supabase.co
SUPABASE_SERVICE_ROLE_KEY: your-key
```
## Adapters
Agentic CMS uses an adapter pattern to support different CMS backends.
### Available
- **Supabase** — For Supabase/PostgreSQL-based CMS setups
### Planned
- **Payload CMS** — TypeScript-native headless CMS
- **Strapi** — Popular open-source headless CMS
- **Directus** — SQL-based headless CMS
- **Custom** — Implement the `CMSAdapter` interface for any backend
### Writing Your Own Adapter
```typescript
import { CMSAdapter } from '@brxce/agentic-cms';
export class MyAdapter implements CMSAdapter {
async listContents(filter?: ContentFilter): Promise<Content[]> {
// Your implementation
}
async getContent(idOrSlug: string): Promise<Content> {
// Your implementation
}
// ... other methods
}
```
## Architecture
```
src/
├── server.ts # MCP server entry point
├── tools/ # MCP tool definitions
│ ├── contents.ts # Content CRUD tools
│ ├── ideas.ts # Idea management tools
│ └── publications.ts # Publication tracking tools
├── adapters/
│ ├── interface.ts # CMSAdapter interface
│ ├── supabase.ts # Supabase adapter
│ └── (future adapters)
└── types.ts # Shared types
```
## Supabase Schema
If you're starting fresh, here's the minimum schema:
```sql
-- Contents
create table contents (
id uuid primary key default gen_random_uuid(),
title text not null,
slug text unique not null,
status text default 'draft' check (status in ('draft', 'review', 'published')),
category text,
body_md text,
tags text[],
hook text,
created_at timestamptz default now(),
updated_at timestamptz default now()
);
-- Ideas
create table ideas (
id uuid primary key default gen_random_uuid(),
raw_text text not null,
source text default 'manual',
promoted_to uuid references contents(id),
created_at timestamptz default now()
);
-- Publications
create table publications (
id uuid primary key default gen_random_uuid(),
content_id uuid references contents(id),
channel text not null,
url text,
published_at timestamptz default now(),
metrics jsonb default '{}'
);
```
## Multi-tenant deployment
agentic-cms 는 **env 기반 multi-tenant** 구조입니다. 같은 코드베이스로 여러 고객을 각자의 Supabase · Storage · 브랜딩으로 서비스할 수 있습니다.
### 시스템 prerequisites (모든 tenant 공통)
```bash
# macOS
brew install ffmpeg yt-dlp node pnpm python@3.12
# Ubuntu/Debian
sudo apt install ffmpeg python3.12 python3.12-venv
pip install yt-dlp
npm install -g pnpm
# Node.js 22+: https://nodejs.org/
```
### 새 고객 onboarding 체크리스트
1. **Supabase 프로젝트 생성** (고객 전용)
- Free/Pro plan, project_ref 기록
- `supabase/migrations/*.sql` 전부 적용 (Studio SQL Editor, 파일명 오름차순 순서대로)
- 자동 적용 항목:
- 14+ 테이블 생성 (contents/ideas/variants/blog_posts/carousels/video_projects 등)
- 5 storage bucket 생성 (content-media, studio-renders, references, finished, blog-images) — migration `20260419000000` 이 자동 처리
- RLS policies 공개 읽기 + service_role 관리
2. **`.env` 파일 3개 작성** (각 프로젝트 루트의 `.env.example` 기준)
- `./.env` — MCP 서버용
- `./dashboard/.env.local` — Next.js dashboard
- `./editor/.env` — Python 영상 편집 서버
3. **Multi-tenant 핵심 env (반드시 고객별로 교체)**
- `SUPABASE_URL` / `SUPABASE_SERVICE_ROLE_KEY` (혹은 `SUPABASE_SERVICE_KEY`)
- `NEXT_PUBLIC_SITE_URL` — 고객 웹사이트 URL
- `NEXT_PUBLIC_BRAND_NAME` — 뉴스레터 헤더 · meta title suffix 에 노출
- `NEXT_PUBLIC_BRAND_HANDLE` / `NEXT_PUBLIC_BRAND_EMOJI` / `NEXT_PUBLIC_BRAND_AVATAR_URL` — 캐러셀 워터마크/아바타
- `NEXT_PUBLIC_CONTACT_EMAIL` / `NEXT_PUBLIC_CONTACT_DOMAIN` — 캐러셀 CTA 슬라이드
- `ANALYTICS_OWN_DOMAINS` — self-referrer whitelist (쉼표 구분)
- `ANALYTICS_VERCEL_KEYWORDS` — vercel preview 도메인 자사 식별
- `NEWSLETTER_FROM` — 뉴스레터 발신인 `"Display Name <addr@domain>"`
- `META_TITLE_SUFFIX` — blog post meta_title 꼬리
- `TABLE_PROJECTS=video_projects` (editor/.env, 필수)
- `STORAGE_MODE=cloud` (editor/.env, 권장)
4. **외부 API Key** (선택 기능)
- `RESEND_API_KEY` — 뉴스레터 발송
- `GOOGLE_SERVICE_ACCOUNT_KEY` — GA4/GSC analytics
- `POSTIZ_API_URL` + `POSTIZ_API_KEY` — 소셜 채널 발행
5. **로컬 개발 기동**
```bash
# MCP 서버용 의존성
npm install && npm run build
# dashboard 의존성
cd dashboard && npm install && cd ..
# editor Python 가상환경
cd editor && python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt && cd ..
# editor Remotion 의존성
cd editor/remotion && pnpm install && cd ../..
# Next.js dashboard + Python editor 동시 기동
cd dashboard && npm run dev:all
```
6. **Claude Code 에서 MCP 서버 연결**
- `.mcp.json` 을 repo 루트에 생성 (`AGENTS.md` 예시 참고)
- Claude Code 재시작 → 43+ MCP 도구 자동 노출
### 고객 분리 원칙
- **DB 분리** — 고객마다 별도 Supabase 프로젝트 (데이터 완전 격리)
- **코드 공유** — 동일 git branch, env 주입만 다름
- **브랜딩 격리** — 위 env 교체만으로 로고·이메일·도메인 전부 고객 것으로 전환
- **배포 분리** — 고객마다 별도 Vercel/Cloud Run 배포 권장 (env 분리 확실)
### 코드에 하드코딩된 브랜딩 없음 원칙
신규 기능 추가 시 "AWC", "agenticworkflows.club", "특정 이메일" 등 구체 값을 직접 박지 말고 env 로 주입. 테넌트별 차이가 생길 여지는 전부 env 통로를 둔다.
---
## Philosophy
> "Nobody cares about your tech. They care that their problem got solved."
Agentic CMS isn't about adding AI to your CMS. It's about making your content workflow AI-native — so agents handle the repetitive work and humans focus on judgment and creativity.
Built by [IntelliEffect](https://intellieffect.com) as part of the [Agentic Workflow Club](https://agenticworkflows.club).
## Contributing
Contributions welcome! Please read [CONTRIBUTING.md](CONTRIBUTING.md) before submitting PRs.
## License
[MIT](LICENSE)
TDQS
Scored across 52 tools
Most tools target a distinct resource+action, and pipeline-step labels in descriptions aid selection. However, the 'media' family (list_media, list_gallery_media, list_videos) and gallery cover tools (set_gallery_cover vs attach_gallery_media role=cover) have overlapping boundaries that require careful reading.
Every tool uses snake_case with a consistent verb_noun pattern (list_*, get_*, create_*, update_*, delete_*, send_*). A few longer names like create_blog_post_from_markdown and link_video_project_to_variant remain readable and follow the same convention.
52 tools is well beyond the 25-tool threshold for a heavy server and risks overwhelming an agent's selection space. Although the server spans many subdomains (blog, gallery, video, carousel, social publishing, pipeline), the surface is bloated and could be split or consolidated.
Core pipeline and CRUD flows are well covered: ideas (list/get/create/update/promote), gallery items (full CRUD), carousels, contents, variants, and publishing. Minor gaps exist, notably missing delete/update operations for blog posts, carousels, contents, variants, media, and video projects, though agents can generally work around these.