Agentic writer

Agentic writer is a Mastra-powered content pipeline: write an article from operating instructions (and an optional author draft), then generate social posts — with human approval on articles before promotion.

flowchart LR
    you([You]) --> notes[Notes] --> writer([Writer workflow]) --> article[Article] --> social([Social workflow]) --> posts[Posts on disk]

    style writer fill:#dbeafe,stroke:#2563eb,stroke-width:2px,color:#1e3a8a
    style social fill:#dcfce7,stroke:#16a34a,stroke-width:2px,color:#14532d

Loading

Cost disclaimer: I am not responsible for unexpected costs when using these agents and workflows.

Getting Started

Set your API keys in .env:

cp .env.example .env
# then edit .env and set OPENAI_API_KEY and DEEPSEEK_API_KEY

Models are tiered for cost vs quality in src/mastra/config/models.ts. Override any agent via env for A/B testing — see docs/agents.md#ab-testing-model-overrides.

Customize the pipeline for yourself (gitignored, local only):

cp src/mastra/config/user-profile.local.example.json src/mastra/config/user-profile.local.json
# edit name, mission, audience, voice, and goals

If user-profile.local.json is missing, the committed example profile (Marty McFly) is used.

Approved articles are saved under data/articles/<article-id>/ (gitignored), with numbered drafts and editor reviews kept alongside approved.md. Social campaigns are written under each article's social/ folder. The social media workflow shows approved articles in an article dropdown after you run the article workflow.

Start the development server:

npm run dev

Open http://localhost:4111 in Studio: run articleWorkflow with your notes (and optional author draft), then socialMediaWorkflow — pick a saved article from the dropdown and choose target platforms.

Or drive the same pipelines from an MCP client while the server is running — see docs/mcp.md (http://localhost:4111/api/mcp/agentic-writer/mcp).

You can start editing files inside the src/mastra directory. The development server will automatically reload whenever you make changes.

Workflows

Workflow	Description
Article workflow	Turns author operating instructions (and an optional author draft) into a researched, written, and human-approved Markdown article.
Social media workflow	Turns the approved Markdown article into platform-native posts and a hero image, saved to disk under the article folder.

See docs/workflows.md for steps, inputs/outputs, and integration details.

MCP

While npm run dev is running, an MCP server exposes start/list/approve tools for articles and social campaigns at http://localhost:4111/api/mcp/agentic-writer/mcp. See docs/mcp.md.

Agents

Six specialized agents power the two workflows. Each agent's tone and personality is centralized in src/mastra/config/personalities.ts so it can be tuned project-wide without touching the agent definitions.

Agent	Model	Description
Researcher	deepseek/deepseek-v4-flash	Extracts topics from operating instructions; when instructions include URLs, fetches those sources only; otherwise searches the web. Produces a research brief for the Writer.
Writer	openai/gpt-5	Drafts and revises the article from the research brief and instructions; develops an optional author draft. Loads technical-essay skills from skills.sh.
Editor	openai/gpt-5-mini	Reviews each draft against instruction intent, research, and optional author draft; returns structured { ready, review } for automatic polish (up to 3 passes) before human approval. Loads edit-article / copy-editing skills.
Strategist	openai/gpt-5-nano	Decides per-platform publication strategy — hook, call to action, and timing — for a social campaign.
Content Creator	openai/gpt-5-mini	Writes platform-native posts (runs in parallel with image generation); loads social / hook skills.
Graphic Designer	openai/gpt-4.1-nano	Creates one on-brand hero image from the article title only (simple schematic figures allowed; no text).

See docs/agents.md for models, tools, full instructions, and personality details.

See docs/observability-memory-and-token-limiter.md for observability, observational memory, and input token limiting.

See docs/customization.md for local profile and article storage.

---

This tool is developed with Mastra, an open-source TypeScript framework for building AI agents and workflows.