magic-code

Open-source TUI agentic AI coding agent. Built in Rust. Fast. Multi-provider.

┌─ magic-code ████████░░ 62% ──────────────────────────────┐
│                                                           │
│ › Fix the failing test in src/auth.rs                     │
│                                                           │
│   ⚙ tool: grep_search                                    │
│   ⚙ tool: read_file                                      │
│   ⚙ tool: edit_file                                      │
│   ⚙ tool: bash (streaming output)                        │
│     running tests...                                      │
│     test auth::test_login ... ok                          │
│     test auth::test_token ... ok                          │
│                                                           │
│ Fixed. The issue was a missing lifetime bound on line 42. │
│ Tests pass now: 47/47 ✓                                   │
│                                                           │
├───────────────────────────────────────────────────────────┤
│ Input                                                     │
├───────────────────────────────────────────────────────────┤
│ claude-sonnet │ ctx ████████░░ 62% │ 12K↓ 2K↑ $0.066     │
└───────────────────────────────────────────────────────────┘

Install

Quick install (Linux/macOS)

curl -fsSL https://raw.githubusercontent.com/kienbui1995/mc-code/main/install.sh | sh

Download binary

Pre-built binaries for Linux (x86_64, aarch64) and macOS (x86_64, aarch64):

# Latest release
gh release download --repo kienbui1995/mc-code --pattern '*linux-x86_64*'
tar xzf magic-code-linux-x86_64.tar.gz
sudo mv magic-code /usr/local/bin/

Build from source

git clone https://github.com/kienbui1995/mc-code.git
cd mc-code/mc
cargo install --path crates/mc-cli

Setup

export ANTHROPIC_API_KEY="your-key"   # or OPENAI_API_KEY, GEMINI_API_KEY, etc.
magic-code                             # start TUI
magic-code "fix the bug in auth.rs"   # single-shot mode

Features

Multi-Provider

Works with 15 providers: Anthropic, OpenAI, Gemini, Groq, DeepSeek, Mistral, xAI, OpenRouter, Together, Perplexity, Cohere, Cerebras, Ollama, LM Studio, llama.cpp. Switch mid-session with /model.

30 Built-in Tools

Tool	Description
bash	Execute shell commands (streaming output)
read_file	Read files with offset/limit
write_file	Create or overwrite files
edit_file	Surgical text replacement with diff preview
batch_edit	Multiple file edits in one call
apply_patch	Apply unified diff patches
glob_search	Find files by pattern
grep_search	Search file contents with regex
subagent	Delegate tasks to isolated sub-conversations
memory_read	Read persistent project facts
memory_write	Save facts across sessions
web_fetch	Fetch URL content (streaming)
web_search	Search DuckDuckGo
lsp_query	Code intelligence (go-to-def, references)
task_create	Spawn background shell commands
task_get/list/stop	Manage background tasks
todo_write	LLM-managed TODO list
ask_user	Ask user for clarification
sleep	Pause execution
notebook_edit	Edit Jupyter notebook cells
worktree_enter/exit	Isolated git branch work
mcp_list/read_resources	MCP server resources

TUI

• Syntax highlighting (syntect)
• Markdown rendering
• Scroll (PageUp/PageDown, mouse wheel)
• Input history (persisted)
• Tab completion for slash commands
• Context window usage bar
• Permission prompts for destructive operations

Intelligence

• True async streaming (tokens render as they arrive)
• LLM-based smart context compaction
• Extended thinking / reasoning blocks
• Image support (/image)
• Long-term memory (persists across sessions)
• @file mentions (auto-read file content)
• Conversation branching (fork/switch)
• Parallel tool execution (up to 4 concurrent)
• Tool result caching (30s TTL for read-only tools)
• Prompt caching (Anthropic, up to 90% cost savings)
• Dynamic token budget
• Mid-stream retry with exponential backoff

Git Integration

• /diff — show changes
• /commit — LLM-generated commit messages
• /log — recent history
• /stash / /stash pop

Safety

• Permission modes: read-only, workspace-write, full-access
• File protection (.env, *.key, .git/* + configurable)
• Workspace sandboxing
• Audit log for all tool executions
• Dry-run mode
• Undo/rollback (/undo)

Developer Experience

• /model — switch provider/model mid-session
• /cost / /cost --total — per-turn breakdown and all-time cost tracking
• /doctor — check connectivity, config, API keys
• /init — project setup wizard
• /export — export conversation to markdown or JSON
• /diff-preview — approve/reject file changes with diff before writing
• /auto-test — auto-run tests after code changes, retry on failure
• /search — search across saved sessions
• /summary — session statistics
• Session save/load/resume
• Pipe mode (echo "fix this" | magic-code)
• Config layering: global → project → local
• Custom instructions (.magic-code/instructions.md)
• Pre/post tool call hooks
• MCP server support
• Plugin system (custom tools from scripts)

Plugins

Create custom tools by adding scripts to .magic-code/tools/:

# .magic-code/tools/deploy.sh
# Deploy to staging environment
#!/bin/sh
echo "Deploying: $PLUGIN_INPUT"
./scripts/deploy.sh staging

# .magic-code/tools/lint.py
# Run linter on a file
import os, subprocess
result = subprocess.run(["ruff", "check", os.environ["PLUGIN_INPUT"]], capture_output=True, text=True)
print(result.stdout or "No issues found")

Supports .sh, .py, .js. Auto-discovered on startup as plugin_<name> tools.

Configuration

magic-code --init  # or /init in TUI

Creates .magic-code/config.toml:

model = "claude-sonnet-4-20250514"
provider = "anthropic"
permission_mode = "workspace-write"

[model_aliases]
fast = "claude-haiku"
smart = "claude-sonnet-4-20250514"

Global config: ~/.config/magic-code/config.toml

Slash Commands

Command	Description
/help	Show all commands
/quit	Exit
/status	Session info
/cost	Session cost (--total for all-time)
/model <name>	Switch model
/diff	Git diff
/commit	Auto-commit with LLM message
/log	Git log
/stash	Git stash (pop to restore)
/compact	Compress context
/undo	Revert last turn's file changes
/save <name>	Save session
/load <name>	Load session
/export	Export to markdown (json for JSON)
/search <q>	Search sessions
/summary	Session stats
/plan	Toggle plan mode
/image	Attach image
/diff-preview	Toggle diff approval for writes
/auto-test	Toggle auto-test after code changes
/clear	Clear output
/init	Project setup
/doctor	Health check
/dry-run	Toggle dry-run

Requirements

• Rust 1.75+ (build from source)
• API key for at least one provider (ANTHROPIC_API_KEY, OPENAI_API_KEY, etc.)

License

MIT