DevFlow AI Agent

Automating Developer Workflows with Claude and MCP

GitHub Repository: github.com/amir-aghdam/dev-flow

Overview

Software development teams today rely heavily on pull requests and continuous integration pipelines to maintain code quality and coordinate collaboration. Yet these processes can be tedious, inconsistent, and error-prone. That’s where DevFlow AI comes in.

DevFlow AI is an intelligent automation agent built using the Model Context Protocol (MCP) and powered by Claude Code. It assists developers by:

  • Analyzing code changes
  • Suggesting PR templates and summaries
  • Monitoring GitHub Actions workflows
  • Notifying teams in Slack

This project is a practical example of how MCP primitives—Tools, Resources, and Prompts—can work in harmony to streamline engineering workflows.

Why DevFlow AI?

Too often, dev teams face issues like:

  • Poorly described pull requests
  • Ignored or misunderstood CI failures
  • Delayed communication between teams

DevFlow AI automates these pain points. It’s not just about saving time—it’s about reducing friction, increasing consistency, and creating room for deeper engineering focus.

Architecture At a Glance

Developer → Git Push → GitHub → Webhook Server
                                  ↓
      Claude Code ← MCP Server ← github_events.json
            ↓
    Slack Notification
Component Purpose
server.py Hosts all MCP Tools and Prompts
webhook_server.py Listens for GitHub workflow events
Slack integration Sends messages based on workflow triggers
Prompt templates Ensure consistent outputs from Claude

The Power of MCP

At the heart of DevFlow AI is MCP, a protocol that defines how large language models like Claude can safely and systematically interact with your tools and data.

🔧 Tools

Tools in DevFlow AI include:

  • analyze_file_changes: Parses diffs and summarizes changes
  • get_pr_templates: Retrieves markdown PR templates
  • send_slack_notification: Delivers formatted updates to Slack

✍️ Prompts

Prompts are where the intelligence shines. Each prompt standardizes tasks like:

  • Analyzing failed CI pipelines
  • Suggesting next actions
  • Drafting team-wide Slack messages

By separating decision logic from raw data, MCP enables Claude to reason clearly with minimal hard-coding.

Key Features

🧠 Smart Pull Request Assistance

Claude uses Tools and Resources to:

  • Identify the nature of the code change
  • Select an appropriate PR template (bug, feature, docs, refactor)
  • Auto-fill the summary based on commit analysis and coding standards

🔍 Real-Time CI/CD Monitoring

When GitHub Actions run:

  • Webhooks are captured and logged
  • Prompts like “Analyze CI Results” summarize failures and suggest next steps
  • Slack messages are formatted and delivered based on team escalation policies

📣 Automated Slack Communication

DevFlow AI sends:

  • Build failure alerts
  • Deployment summaries
  • Code review reminders

All notifications are tailored using team resource data and Slack channel templates.

Technical Stack

  • Python 3.11+
  • FastMCP for building the MCP server
  • aiohttp for webhook listening
  • Cloudflare Tunnel for public GitHub webhook access
  • Claude Code for reasoning and formatting
  • Slack Webhooks for outbound messaging

Demo Scenarios

  • Developer pushes a commit → DevFlow AI analyzes changes, suggests PR template
  • CI pipeline fails → Claude summarizes the root cause, creates Slack update

“Hey Claude, summarize the build failures and notify the related team.” ✅ Done.

Why MCP Matters

Most automation tools today require tight coupling between logic and data. MCP flips this model:

  • Claude reads Resources like docs and guidelines
  • Claude uses Tools to gather structured info
  • Claude applies Prompts to make intelligent, human-like decisions

The result is a system that’s not only powerful, but also maintainable and extensible.

✨ Built with MCP + Claude Code. Designed for real-world developer impact.