jdbg gives AI coding agents a real Java debugger

jdbg is a debugger interface built so can inspect a running Java program directly. The problem it targets is familiar in coding: agents often guess from error traces, add many temporary print statements, or repeat the same failing command without learning what is happening inside the program. jdbg wraps the JDK’s built-in jdb and makes it easier for agents to control.

It supports persistent debug sessions, breakpoints, stepping, local variables, stack frames, watches, threads, and locks. It includes MCP tools and setup paths for Claude Code, Codex, and OpenCode. It is designed to work without shell-script glue or s, with Windows as the first target while still aiming to be .

The main idea is that an agent should use runtime facts from the JVM instead of guessing the state of a Java app from logs alone.

Key points

  • jdbg lets inspect live Java runtime state instead of only reading logs.
  • It wraps jdb, the debugger included with the JDK, in an agent-friendly interface.
  • It supports breakpoints, stepping, local variables, stack frames, watches, threads, and locks.
  • It provides MCP support for Claude Code, Codex, and OpenCode.
  • Its practical goal is to help agents make better debugging decisions with real runtime facts.

Sources covering this story (2)

Read original