Git to Bluesky AI Agent

An AI-powered automation tool that converts your latest GitHub commit into a concise, developer friendly Bluesky post using an LLM, delivered via Model Context Protocol (MCP).

This project supports two independent modes:

1. Claude / MCP stdio version

for direct MCP compatible agents (like Claude)

2. CLI-based Agent version

a full MCP Client & Server setup using SSE, usable from the terminal

---

What This Project Does

• Fetches the latest commit from a GitHub repository
• Extracts the changed code
• Uses an LLM (Groq – LLaMA 3.3 70B) to generate a short, clean post
• Automatically posts to Bluesky
• Secured using a tool access key

Perfect for:

• Daily coding challenges
• #BuildInPublic posts
• Developer content automation

---

High-Level Architecture

GitHub Repo
   ↓
Fetch Latest Commit
   ↓
Extract Code
   ↓
LLM (Groq API)
   ↓
Generate Post
   ↓
Bluesky API

---

Project Structure

Git_To_Bsky_Agent/
│
├── Claude_version/
│   └── src/
│       └── index.ts        # MCP stdio-based server
│
├── CLI_version/
│   ├── Client/
│   │   └── src/index.ts    # CLI MCP client
│   │
│   └── Server/
│       └── src/index.ts    # MCP SSE server (Express)
│
├── .gitignore
├── .env.example
└── README.md

---

Environment Variables

Create a .env file in both Claude_version and CLI_version (Server & Client) as needed.

.env.example

# LLM
GROQ_API_KEY=your_groq_api_key

# GitHub
GITHUB_OWNER=your_github_username
GITHUB_REPO=your_repository_name

# Bluesky
BLUESKY_HANDLE=your_handle.bsky.social
BLUESKY_APP_PASSWORD=your_app_password

# Security
TOOL_ACCESS_KEY=some_secret_key

---

Version 1: Claude / MCP stdio Version

Description

This version runs as an MCP server over stdio, designed for tools like Claude Desktop or any MCP compatible host.

How to Run

cd Claude_version
npm install
npm run dev

You should see:

MCP Server running on stdio

MCP Tool Available

Tool Name: createBskyPostFromGithub

Inputs:

{
  "day": "12",
  "key": "TOOL_ACCESS_KEY"
}

---

Version 2: CLI-Based MCP Agent (Client & Server)

Description

A complete SSE-based MCP setup:

• Express server exposes MCP tools
• Terminal client chats with an LLM and invokes MCP tools automatically

---

▶️ Start the Server

cd CLI_version/Server
npm install
npm run dev

Server runs on:

http://localhost:4000/sse

---

Start the Client

cd CLI_version/Client
npm install
npm run dev

You’ll see:

Agent: Welcome YourName, how can I assist you today?

Try prompts like:

Create today's Bluesky post
Post my Day 5 update

The agent will automatically invoke the MCP tool.

---

Security Model

• MCP tools are protected via TOOL_ACCESS_KEY
• Unauthorized tool usage is blocked
• All secrets are environment-based

---

Known Limitations

• Assumes at least one file exists in the latest commit
• Uses only the first changed file
• No retry logic for API failures

These can be improved in future iterations, along with the agentic architecture.

---

Future Improvements

• Multi file commit support
• Commit diff summarization
• Post preview mode
• Twitter / LinkedIn support
• Shared utility package
• Agentic Architechture

---

Author

Jayesh Verma

Built as a real world AI automation + MCP learning project.

---

License

MIT License

You’re free to use, modify, and distribute this project.