DEV Community: NISHANT UNAVANE

Atlas MCP Server — 33 Developer Tools in One Package

NISHANT UNAVANE — Thu, 28 May 2026 11:12:22 +0000

Atlas started as a simple code analysis tool and grew into a suite of 33 developer tools packaged as an MCP server. It integrates with Cursor, Windsurf, Claude Desktop, GitHub Copilot, and VS Code.

The core pipeline follows a six-stage process: Intent, Context, Decomposition, Variants, Critique, Optimization. When you ask Atlas to refactor a function, it understands your intent, gathers project context, breaks the task into subtasks, generates multiple approaches, critiques each one, and applies the best solution.

The RAG search feature is technically the most interesting. Instead of treating code as plain text for keyword matching, Atlas builds a knowledge graph from your codebase. It understands relationships between files, functions, imports, and data flow. You can ask "how does authentication work?" and get a contextual answer with relevant code paths.

The ML bug predictor analyzes complexity metrics, churn rates, and historical patterns to forecast issues with 70-85% accuracy. Training required collecting metrics from hundreds of open source projects.

The frontend developer tools include a Performance Doctor that detects React re-render issues, bundle bloat, and memory leaks. The CSS Wizard analyzes specificity conflicts and converts between BEM, Tailwind, and CSS Modules. The Animation Studio generates code for GSAP, Framer Motion, and CSS keyframes.

Atlas works without API keys. If none are provided, it delegates generation to your IDE's built-in AI. This makes it truly zero-configuration.

If you want to see the full source code or read about my other projects, visit my portfolio at nishantunavane.qzz.io.

Check out the Project
View Source Code on GitHub

What developer tool would you add to an IDE if you could build anything? Let me know in the comments!

Building Axiom — A Claude Code-Style Terminal Agent with 7 LLM Providers

NISHANT UNAVANE — Thu, 28 May 2026 11:08:32 +0000

After using Claude Code CLI extensively, I wanted to understand the architecture by building my own version. Axiom is a terminal-based coding agent with a premium TUI that supports seven LLM providers.

The project is a monorepo with five packages: a unified LLM abstraction, an agent runtime with tool execution, a terminal UI framework, a React-based premium TUI, and the interactive coding agent itself.

Building a 60fps terminal UI was harder than expected. Terminals are not designed for smooth animation. I implemented a frame-based renderer with dirty region optimization — only changed parts of the screen are redrawn. This keeps the interface responsive during streaming token-by-token responses.

Vim mode was fun to implement. Normal mode treats h, j, k, l as movement keys. Insert mode passes keystrokes through. The mode switch is instantaneous with clear visual feedback in the status bar.

Security is built into every tool execution. The bash tool checks commands against dangerous patterns before running them. Path validation prevents directory traversal. Read-only filesystem detection prevents accidental writes to system files.

Session management uses JSONL persistence with branching support. You can try multiple approaches to a problem and compare results in parallel branches.

If you want to see the full source code or read about my other projects, visit my portfolio at nishantunavane.qzz.io.

Check out the Project
View Source Code on GitHub

What features would you want in a terminal coding agent? Do you prefer CLI-based AI tools or editor-integrated ones? Let me know in the comments!

Controlling Blender with AI — Building an MCP Server for 3D Creation

NISHANT UNAVANE — Thu, 28 May 2026 11:06:28 +0000

Blender's Python API is powerful but has a steep learning curve. What if you could describe a 3D model in plain language and have an AI build it inside Blender automatically?

That is what Blender MCP does. It is an MCP server that connects AI assistants like Claude and GitHub Copilot to Blender in real-time.

The architecture is simple. The MCP server runs as a Node.js process and spawns Blender as a background subprocess. When you tell the AI to "create an Indian temple with red marble material," the server generates a Python script using Blender's bpy API and executes it in the background Blender instance.

The server includes procedural generation templates for Indian-themed models — temples with detailed pillars and shikharas, auto-rickshaws with functional wheels, traditional thalis with rice and multiple curries, and human figures in traditional attire. Each template is a parameterized Python script that generates geometry procedurally.

Managing the Blender subprocess was the hardest part. Blender takes seconds to start, and long operations can timeout. I built a connection pool that keeps Blender running in the background and automatically reconnects if it crashes.

Material generation uses Blender's node system instead of texture files. Wood grain, marble in multiple colors, metals like gold and copper — each material is a node tree created programmatically.

If you want to see the full source code or read about my other projects, visit my portfolio at nishantunavane.qzz.io.

Check out the Project
View Source Code on GitHub

Have you tried using AI for 3D modeling? What kind of scenes would you generate? Let me know in the comments!

I Built a VS Code Extension That Runs Code in 40+ Languages published: true

NISHANT UNAVANE — Thu, 28 May 2026 11:02:48 +0000

I was tired of switching between terminal tabs to run code snippets. Python here, JavaScript there, Rust somewhere else. So I built a VS Code extension that handles all of them.

The extension is called Code Runner. It adds a run button to your editor that executes the current file using the right compiler or interpreter. It supports 40+ languages out of the box.

The hardest part was abstracting language-specific execution into a unified interface. Each language has different compiler flags, execution patterns, and error formats. Python runs directly. C++ needs compiling first. Rust has cargo. TypeScript needs ts-node or a build step.

I settled on a configuration-driven approach. The extension ships with default profiles for each language. Users can override any of them through VS Code settings. This gives beginners a zero-config experience while letting power users customize everything.

The output panel was another challenge. I wanted syntax-highlighted results with inline error markers. VS Code's output channel is plain text only. I ended up building a custom WebView-based output panel that formats results with proper highlighting.

Error handling is where most code runners fall short. If your Python script throws a traceback, the extension parses it and creates problem markers at the correct lines. You can navigate errors with F8 just like regular compiler errors.

The extension has been downloaded a few thousand times. The most common feedback is that it works when nothing else does. That's the best compliment a developer tool can get.

I learned more about child process management and cross-platform compatibility from this project than from any tutorial. Things like resolving PATH variables differently on Windows, handling spaces in file paths, and managing long-running processes gracefully.

If you want to see the full source code or read about my other projects, visit my portfolio at nishantunavane.qzz.io.

Check out the Project

If you want to see the full source code or contribute, you can find the repository here:

View Source Code on GitHub

(Optional: If your repo is public, use the embed below instead!)

IamNishant51 / CODE-RUNNER-vscode-extension

extension for vs code to run c and c++ programms inside vs code without any online compiler or switching tabs

VS Code Code Runner Extension 🚀

A powerful VS Code extension that allows you to run C, C++, and Python code directly within VS Code without switching between applications or using online compilers.

Features ✨

Multi-language Support: Run C, C++, and Python code seamlessly
Real-time Output: View program output directly in VS Code
Interactive Input: Support for program input during execution
Clean Interface: Modern and intuitive user interface
File Management: Automatic file handling and cleanup
Cross-platform: Works on Windows, Linux, and macOS

Prerequisites 🛠️

Ensure you have the following installed:

Visual Studio Code
For C/C++: GCC/G++ compiler
For Python: Python interpreter
Required extensions dependencies will be automatically installed

Installation 📥

Open VS Code
Go to Extensions (Ctrl+Shift+X)
Search for "Code Runner"
Click Install

Usage 💻

Open your code file in VS Code
Click the "Code Runner" icon in the activity bar
Enter your file…

View on GitHub

I’m curious—how do you usually handle running snippets quickly? Do you stick to the terminal, or have you customized your IDE for specific languages? Let me know in the comments!