How to Build a Marketing Team Using Paperclip and AI Agents (with System Prompts)
Don't want to figure this out alone? I walk members through every step inside the community. Join the Skool → skool.com/raycfu
Paperclip is a Node.js server with a React UI that coordinates multiple AI agents into a structured organization. You bring your own agents (Claude Code, OpenClaw, Codex, or any HTTP-based agent) and Paperclip handles the coordination.
Before Paperclip, running multiple AI agents looked like 10 open terminal tabs with no way to track what any of them were doing, no budget controls, and a total loss of context every time the machine rebooted. Paperclip replaces that with a dashboard, a ticket system, and an audit log.
What you get: org charts with reporting lines, per-agent token budgets that pause automatically when hit, scheduled heartbeats so agents wake up and work without you prompting them, task tracking where every action is logged as an issue with full tracing, governance controls with approval gates for risky actions, and goal alignment so every task traces back to a company mission.
Open source. Self-hosted. No Paperclip account required. GitHub: github.com/paperclipai/paperclip
REQUIREMENTS
Node.js 20 or higher. Check with node --version. Install via nvm if needed. pnpm 9.15 or higher. Install with npm install -g pnpm. Git. An AI provider API key (Anthropic, OpenAI, or any provider you prefer).
STEP 1: INSTALL PAPERCLIP
Run this in your terminal:
npx paperclipai onboard --yes
This starts the API server at http://localhost:3100. An embedded PostgreSQL database is created automatically. No database setup required. The onboard command creates a project scaffold with directories for agent configurations, a shared workspace, and logging.
Open http://localhost:3100 in your browser. You should see the Paperclip dashboard.
STEP 2: DEFINE YOUR COMPANY MISSION
Before you create any agents, set your company mission. This matters because every task your agents work on traces back to this mission. Vague goals produce vague work.
Go to Settings in the Paperclip dashboard. Set your company mission to something specific:
"Generate 14 pieces of content per week across TikTok, Instagram, LinkedIn, and X. All content must be on-brand, based on trending topics in our niche, and reviewed before publishing. Target: grow social following by 20% per month while maintaining consistent posting cadence."
Every agent you create will see this mission. It shapes how they prioritize their work.
STEP 3: CREATE YOUR 4 AGENTS
Paperclip works best with tight, well-scoped agents. One agent per role. Each gets its own system prompt, its own tools, and its own area of responsibility.
AGENT 1: CONTENT STRATEGIST
This agent is the brain of the operation. It finds what is trending, decides what content to create, and writes briefs that the other agents execute.
Role: Research trending topics and plan the content calendar.
System prompt: "You are a Content Strategist for an AI-focused social media brand. Your job is to find trending topics in the AI automation and making money with AI niche, analyze why they are trending, and create content briefs for the production team. Every brief must include: the topic, the hook (under 10 words), the content format (slideshow, video, or carousel), the target platform, the angle, and suggested hashtags. You create 2 to 3 briefs per day. You check trend data from Virlo every morning. You review performance data weekly and adjust your strategy based on what actually performs."
Tools to connect: Virlo (trend research), Notion or Google Sheets (content calendar), access to analytics.
Token budget: Set a monthly limit. Start with $15/month and adjust based on usage.
Heartbeat schedule: Daily at 9:00 AM. The agent wakes up, checks trends, creates briefs, and assigns them.
AGENT 2: VIDEO EDITOR
This agent takes briefs from the Content Strategist and produces the actual content assets.
Role: Generate videos, slideshows, and visual content from briefs.
System prompt: "You are a Video Editor for an AI-focused social media brand. You receive content briefs from the Content Strategist. For each brief, you produce the specified content format. For TikTok slideshows: create 6 slides (hook, problem, 3 value points, CTA) at 1080x1920. For Instagram Reels: create 30-second vertical video. For LinkedIn carousels: create 5 to 8 slides at 1080x1080. Use the Remotion skill for video generation. Every asset must follow the brand guide for colors, fonts, and tone. When finished, send assets to the QA Reviewer before they go anywhere else."
Tools to connect: Remotion (video and slide generation), file system access for saving assets.
Token budget: $20/month (this agent uses more tokens because of content generation).
Heartbeat schedule: Daily at 9:10 AM (10 minutes after the Strategist so briefs are ready).
AGENT 3: SOCIAL MEDIA MANAGER
This agent handles distribution. It takes finished assets, writes platform-specific captions, and schedules everything.
Role: Write captions, schedule posts, and manage the publishing calendar.
System prompt: "You are a Social Media Manager for an AI-focused social media brand. You receive finished content assets from the Video Editor (after QA approval). For each asset, you write platform-specific captions. TikTok captions are casual, use trending sounds references, and stay under 150 characters. Instagram captions are slightly longer with 5 to 8 hashtags. LinkedIn posts are professional and value-driven at 200 to 400 words. X posts are punchy and under 280 characters. You schedule all posts through Postiz at optimal times for each platform. You manage the full weekly calendar and ensure consistent posting cadence of 2 posts per day."
Tools to connect: Postiz (scheduling and publishing), access to TikTok, Instagram, LinkedIn, and X through Postiz integrations.
Token budget: $10/month.
Heartbeat schedule: Daily at 10:00 AM (after QA review is complete).
AGENT 4: QA REVIEWER
This is the agent most people skip and the one that matters most. Nothing gets published without passing QA first.
Role: Review every piece of content before it goes live.
System prompt: "You are a QA Reviewer for an AI-focused social media brand. Every content asset must pass through you before publishing. For each asset, check: video resolution is correct for the target platform, hook appears in the first 2 seconds or first slide, CTA is present on the final slide or in the caption, brand voice matches the brand guide (no corporate language, no AI-sounding phrases, no em dashes), all required hashtags are present, no banned words or phrases, captions are the correct length for each platform. If an asset passes, approve it and notify the Social Media Manager. If an asset fails, flag the specific issue, explain what needs fixing, and send it back to the responsible agent. Never approve content that does not meet all criteria."
Tools to connect: File system access to review assets, notification system to flag issues.
Token budget: $8/month (this agent mostly reads and evaluates, minimal generation).
Heartbeat schedule: Daily at 9:30 AM (after Video Editor finishes, before Social Manager publishes).
STEP 4: SET UP THE DAILY ROUTINE
The power of Paperclip is that once you set up the routine, the team runs itself. Here is the daily flow:
9:00 AM: Content Strategist wakes up. Connects to Virlo. Scans for trending topics in your niche. Creates 2 to 3 content briefs. Assigns them as tasks to the Video Editor.
9:10 AM: Video Editor wakes up. Reads the new briefs. Produces the content assets (slides, videos, carousels). Saves them and notifies QA Reviewer.
9:30 AM: QA Reviewer wakes up. Reviews every asset against the checklist. Approves passing content, sends failures back with specific notes.
10:00 AM: Social Media Manager wakes up. Takes approved assets. Writes platform-specific captions. Schedules everything through Postiz for optimal posting times throughout the day.
This entire cycle takes about 30 to 45 minutes of agent time. You wake up to a dashboard showing what was created, what was approved, what was scheduled, and what needs your attention (if anything).
STEP 5: SET UP THE WEEKLY FEEDBACK LOOP
This is the step that makes the content better over time instead of staying flat.
Create a weekly routine that runs every Monday at 8:00 AM (before the daily routine starts). The Content Strategist reviews the past week's performance data: which posts got the most views, which got the most engagement, which flopped.
But here is the key part: you go through each post yourself and add your own comments on what worked and what did not. The AI uses your real feedback to adjust what it creates next week. Without this feedback loop, the agents keep producing at the same quality level forever. With it, every week gets better because the strategy is informed by actual results plus your judgment.
Add these comments directly in Paperclip's task system. The Content Strategist reads them on its next heartbeat and incorporates them into the following week's briefs.
STEP 6: AUTOMATE PRODUCT UPDATES (THE RETENTION PLAY)
This is the part most people miss. Connect your GitHub repo to the Paperclip workspace. Now every time your team ships a feature or pushes an update, one of your agents automatically writes a product update, formats it for each channel (Discord, Slack, Telegram, email newsletter), and schedules it as a draft.
Your users actually find out about the stuff you build for them. Retention is always cheaper than acquisition and this runs without you writing a single update post.
To set this up: add a GitHub webhook that triggers when a PR is merged into your main branch. The webhook fires a task to the Content Strategist who reads the PR description, writes a user-facing update, and passes it through the normal QA and scheduling pipeline.
STEP 7: MONITOR COSTS AND PERFORMANCE
Paperclip tracks spending per agent automatically. Check the dashboard weekly. You should see:
Content Strategist: $10 to $15/month Video Editor: $15 to $25/month Social Media Manager: $8 to $12/month QA Reviewer: $5 to $10/month Total: $38 to $62/month for a full marketing team producing 14+ posts per week
If any agent is burning through tokens faster than expected, check its task logs. Usually the issue is either a prompt that is too verbose, an agent stuck in a retry loop, or unnecessary context loading. Tighten the prompt and the costs come down.
Compare this to hiring: a social media manager costs $3,000 to $5,000/month. A content creator costs $2,000 to $4,000/month. A video editor costs $2,000 to $3,000/month. You are replacing $7,000 to $12,000/month in labor with $40 to $60/month in API costs.
TIPS FOR GETTING THE BEST RESULTS
Allocate models by role. Use Claude Opus or GPT-5.4 for the Content Strategist since it needs the most reasoning. Use Sonnet or GPT-5.4-mini for the QA Reviewer since it is mostly checking against criteria. Match model capability to task complexity, not to the agent's perceived importance.
Keep agent scopes tight. One agent, one job. The moment an agent tries to do two different things (research and production, or writing and scheduling), quality drops. Specialization is the whole point.
Start with one platform. Get the pipeline working for TikTok only. Then add Instagram. Then LinkedIn. Then X. Adding all four at once before the pipeline is stable creates debugging nightmares.
Use file-based communication between agents. Agents write outputs to a shared directory and other agents read from that directory. This is easier to inspect and debug than message passing. You can open the shared workspace at any point and read exactly what each agent produced.
Review agent work for the first 2 weeks. Do not set up the full automation and walk away on day one. Check every output for the first 14 days. Adjust prompts based on what you see. Once the quality is consistent, reduce your oversight to the weekly feedback loop.
Set token budgets conservatively. You can always increase them. An agent that runs out of budget pauses automatically in Paperclip. An agent without a budget that gets stuck in a loop can burn through hundreds of dollars overnight.
Document your agent prompts in a central file. When you improve a prompt, note what you changed and why. Three months from now you will not remember why you added a specific instruction unless you wrote it down.
SCALING BEYOND MARKETING
Once the marketing team is running, you can add agents for other functions using the same pattern. A customer support agent that monitors your help channels and responds to common questions. A sales agent that qualifies inbound leads from your content. A research agent that monitors competitor activity and reports weekly. A development agent that handles bug triage and simple fixes.
Each new agent adds to the org chart. Each one gets its own scope, budget, and heartbeat. Paperclip coordinates all of them. One person managing a 10-agent Paperclip company can produce the output of a 15-person team. That is not a pitch. That is what early adopters are already reporting.
The businesses that figure out agent orchestration first will have a compounding advantage. The window is open right now and very few people are through it.
Install: npx paperclipai onboard --yes Dashboard: http://localhost:3100 GitHub: github.com/paperclipai/paperclip Docs: paperclip.ing
You just read the full playbook. Most people will close this tab and never implement it. The ones who do usually hit a wall around the technical setup and quit.
Inside the Skool, I walk you through the exact build step-by-step, troubleshoot your setup live in the community, and share the scripts and templates I use to actually land paying clients.
If you want the shortcut instead of the long way around: