Building kdex: How I Built a Complete CLI Tool Using GitHub Copilot CLI

This is a submission for the GitHub Copilot CLI Challenge

What I Built

I built kdex—a full-featured Rust CLI for indexing code, docs, and notes for AI-powered search—entirely with GitHub Copilot CLI. From initial brainstorming to a production-ready tool published on crates.io with CI/CD, landing page, and MCP server integration. Here's how it went.

As developers, we accumulate knowledge everywhere: code across dozens of repositories, notes in Obsidian vaults, documentation scattered in wikis. When working with AI assistants, they have no idea what's in your files—you end up copy-pasting context manually.

I wanted to build a tool that solves this. A local-first CLI that indexes everything and lets AI assistants search it via the Model Context Protocol (MCP).

Demo

Website: https://urbanisierung.github.io/kdex/
Github: https://github.com/urbanisierung/kdex

My Experience with GitHub Copilot CLI

The Approach: Brainstorm → Roadmap → Execute

Phase 1: Requirements Brainstorming

I started by having a conversation with Copilot CLI about what I wanted to build. We discussed:

• Target users (developers, knowledge workers, AI-assisted workflows)
• Core features (local indexing, full-text search, MCP integration)
• Tech stack decisions (Rust, SQLite with FTS5, ratatui for TUI)
• UX considerations (dual-mode CLI: interactive TUI + traditional commands)

The result was a comprehensive .github/copilot-instructions.md that established all the ground rules: coding standards, documentation requirements, folder structure, and project conventions.

Phase 2: Roadmap Generation

Next, I asked Copilot CLI to create a detailed implementation roadmap. It generated a 73KB doc/roadmap.md with:

• 14 implementation phases
• Each phase broken into parts with specific action items
• Code examples, SQL schemas, and module structures
• User journeys for different personas
• UX specifications including edge cases

The roadmap became the source of truth. As features were completed, Copilot CLI would check off items and move to the next.

Phase 3: Implementation

With a clear roadmap, I gave a simple instruction: "Implement the next phase."

And it did. Phase after phase.

---

The Implementation Journey

Foundation (Day 1)

• Scaffolded Rust project with proper structure
• Implemented configuration system with OS-aware paths
• Created SQLite database layer with FTS5 full-text search
• Built the indexer with binary file detection and .gitignore support
• Developed CLI commands: index, search, list, update, remove

TUI & Polish (Day 2)

• Built full-screen interactive terminal UI with ratatui
• Added real-time search, file preview, repository management
• Implemented loading states, confirmation dialogs, welcome screen
• Fixed clippy warnings (the project maintains a zero-warning policy)

MCP Server (Day 3)

• Integrated the rmcp crate for MCP protocol
• Implemented 4 tools: search, list_repos, get_file, get_context
• Created integration documentation for GitHub Copilot, Claude, and Gemini

Vector Search (Day 4)

• Added semantic search using fastembed (MiniLM embeddings)
• Implemented hybrid search with Reciprocal Rank Fusion
• Built text chunking for large files

Remote Repositories (Day 5)

• Added ability to index GitHub repos by URL
• Implemented background sync with auto-pull
• Added SSH and token authentication for private repos
• Built config export/import for machine migration

CI/CD & Release (Day 6)

• Set up GitHub Actions for testing on Linux/macOS/Windows
• Created release workflow with cross-platform binary builds
• Configured crates.io publishing with Trusted Publishing (OIDC)
• Built install script for curl-based installation

Marketing & Documentation (Day 7)

• Rewrote README with engaging marketing content
• Created landing page for GitHub Pages
• Documented Ollama integration for local LLMs
• Added add-mcp and self-update commands