coding-agent

A terminal-based AI coding assistant that reads, writes, and runs code through tool calls. Point it at any OpenAI-compatible API and it becomes a pair programmer that can explore your codebase, edit files, run commands, and search the web.

Quick start

# Clone and install
git clone https://github.com/Kiran-Jones/coding-agent.git
cd coding-agent
cp .env.example .env   # add your API key and endpoint
make install

# Run
coding-agent

What it can do

The agent has eight built-in tools it calls autonomously:

• Read & write files with automatic undo/redo snapshots
• Run terminal commands with streaming output
• Search the web and read webpages via DuckDuckGo
• Git operations with command-chaining protection
• Planning mode (/plan) for read-only exploration before making changes

It also supports MCP servers for extending its capabilities.

Slash commands

Command	Description
/plan	Enter planning mode (read-only exploration)
/new	Start a new session
/sessions	List saved sessions
/load /delete	Manage sessions
/model	Switch model
/undo /redo	Revert or replay file changes
/history	View snapshot history
/mcp	Show MCP server status
/verbose	Toggle detailed tool output
/usage	Show token counts

File mentions

Reference files directly in your prompt with @:

@src/agent.py refactor the _parse_stream method
@utils.py:10-25 explain these lines

Configuration

Environment (.env):

API_KEY=your-api-key
ENDPOINT_URL=http://localhost:8000/v1/chat/completions
MODEL_NAME=your-model

Agent context: Drop an AGENT.md file in your project root (or ~/.coding-agent/AGENT.md globally) to give the agent persistent instructions.

MCP servers: Add a mcp_config.json to your project or ~/.coding-agent/mcp_config.json globally. See mcp_config.example.json for the format.

Development

make install       # Install deps (including dev tools)
make check         # Lint + format check + type check
make test          # Run all tests
make format        # Auto-format code

Architecture

src/coding_agent/
  main.py              CLI, REPL, slash commands
  agent.py             Core agent loop and tool dispatch
  tools.py             Built-in tools with auto-generated schemas
  mcp_manager.py       MCP server lifecycle and tool routing
  session_manager.py   Conversation persistence
  snapshot_manager.py  File change tracking (undo/redo)
  memory_utils.py      Context window compaction
  agent_ui.py          Streaming display and approval prompts
  markdown_renderer.py Syntax-highlighted markdown output
  status_bar.py        Terminal status line

License