magikarp

An open-source coding assistant CLI tool built with Go. Magikarp provides an interactive terminal interface for AI-powered coding assistance with support for multiple LLM providers (Claude, GPT, Gemini, Mistral).

This project's journey will be document over on Subtack as part of my Build Your Own Claude Code series. Stay tuned!

Prerequisites

Before installing Magikarp, ensure you have the following installed:

• Go 1.24.2 or later - Download Go
• Git - Install Git
• Make (optional, but recommended) - Usually pre-installed on macOS/Linux, install on Windows

Quickstart

Local Development Setup

1. Clone the repository:

   git clone https://github.com/pprunty/magikarp.git
   cd magikarp

2. Install dependencies and build:

   make install

3. Run Magikarp:

   make run

Alternative Commands

• Build only: make build (creates bin/magikarp)
• Clean build artifacts: make clean
• Format code: make fmt
• View all commands: make help

Configuration

Set up your API keys by creating a .env file:

1. Copy the template:

   cp .env.example .env

2. Edit .env with your API keys:

   # Anthropic API Key (for Claude models)
   ANTHROPIC_API_KEY=your_anthropic_api_key_here
   
   # OpenAI API Key (for GPT models)  
   OPENAI_API_KEY=your_openai_api_key_here
   
   # Google Gemini API Key (for Gemini models)
   GEMINI_API_KEY=your_gemini_api_key_here
   
   # Mistral AI API Key 
   MISTRAL_API_KEY=your_mistral_api_key_here

Alternatively, you can set up your API keys as environment variables:

export ANTHROPIC_API_KEY="your-anthropic-key"
export OPENAI_API_KEY="your-openai-key"
export GEMINI_API_KEY="your-gemini-key"
export MISTRAL_API_KEY="your-mistral-key"

Where to get your API keys

• Anthropic (Claude): https://console.anthropic.com/account/keys
• OpenAI (GPT): https://platform.openai.com/account/api-keys
• Google Gemini (AI Studio): https://makersuite.google.com/app/apikey (create an API key in Google AI Studio)
• Mistral AI: https://console.mistral.ai/api-keys
• Alibaba: https://www.alibabacloud.com/help/en/model-studio/first-api-call-to-qwen

Feature Checklist

• Basic UI terminal
• LLM Provider SDK with sbtract interface for Gemini, GPT, Claude, also include claude-code SDK
• Optimized tools (i.e find_file) with toolboxing (i.e filesystem/ -> read_file, write_file, update_file etc.)
• Release first version for brew, yum, go install, etc. using GoReleaser
• Speech-to-text mode
• /init command for creating AGENT.md (hopefully LLM providers can agree on universal convention sometime soon...)
• Automatic model selection based on user prompt (i.e auto choose best model for the task)
• MCP integration
• Default built-in magikarp agent which on user prompt writes back to core codebase to do things like create new slash commands, new tools, etc.
• Show file difference on update
  • Prompt user to accept/reject updates
• Parallel processing for independent tasks
  • master/slave toggle for spinning up multiple agents to complete long tasks
• Plan mode with optimized model selection, i.e setup could use o3 for planning, gemini for scrutinizing, claude for coding up the plan
• Ability to write back to system prompt with hot reload, something like Karpathy mentions here
• Run magikarp as shorcut on terminal, i.e mgk --prompt push all changes to git

Contributing

If you have any ideas on what else you'd like to see in this project, don't hesitate to let me know on X, or by creating an issue on GitHub.

License

Licensed under the MIT license.

Copyright (c) 2025 Patrick Prunty