langchain-ai/social-media-agent

Social Media Agent – What It Does

Purpose: An end‑to‑end “agent” that takes a web URL, extracts the page’s content, and automatically drafts a Twitter and LinkedIn post for you. The flow is human‑in‑the‑loop: after the LLM generates a draft you can edit, approve, or reject it before it is actually published.

Key Components

• LLM backend – Uses Anthropic’s Claude (or any Anthropic model) via the Anthropic API to write the posts.
• Web scraping – FireCrawl fetches the article, blog, GitHub repo, YouTube transcript, etc.
• Auth & posting – The Arcade service (or your own Twitter/LinkedIn developer credentials) handles OAuth, token storage and the actual API calls to publish.
• State & orchestration – A LangGraph graph runs the whole pipeline (scrape → summarize → draft → human review → post). The graph is served locally with the langgraph-cli and can be deployed on LangGraph Cloud.
• Optional integrations – Supabase for image storage, Slack for ingesting URLs via a channel, and GitHub for repo‑specific content.

How to Run (quick‑start)

1. Clone the repo and install dependencies (yarn install).
2. Copy .env.quickstart.example → .env and fill in the required keys (Anthropic, FireCrawl, Arcade, LangSmith optional).
3. Install the LangGraph CLI (pip install langgraph-cli).
4. Start the in‑memory LangGraph server: yarn langgraph:in_mem:up.
5. Run yarn generate_post – the script calls the generate_post graph on a sample LangChain blog URL. Results appear in LangSmith or in the Agent Inbox UI.

Full‑feature setup adds:

• YouTube video handling via Google Vertex AI.
• Image selection/upload via Supabase.
• Slack‑driven cron jobs that pull URLs from a channel and automatically invoke the graph.
• Direct Twitter/LinkedIn OAuth (instead of Arcade) for full media upload support.
• GitHub URL parsing and organization‑level LinkedIn posting.

Customization

• Prompt files under src/agents/generate-post/prompts/ let you change business context, example posts, structure instructions, and style guidelines.
• Post‑style prompts control tone and formatting without touching the code.

Typical Use‑Case A marketing team drops a link into a designated Slack channel. A daily cron runs, the agent scrapes the link, drafts a concise, brand‑aligned tweet and LinkedIn update, shows the draft in the Agent Inbox for a quick edit, and then publishes it automatically.

Who Might Want This

• Content marketers looking to automate social‑media copy generation.
• Developers who want a ready‑made LangGraph example of a HITL workflow.
• Teams that already use LangChain/LangGraph and want to extend it to social platforms.

---

Bottom Line: The Social Media Agent is a genuine, runnable project that demonstrates how to combine LLM generation, web scraping, OAuth authentication, and a human‑in‑the‑loop UI to automate posting on Twitter and LinkedIn.