Documentation

Start at the top and go as deep as you need: getting started takes five minutes, the guides cover every task, the reference is generated from the binary itself, and the advanced section is the full design.

1Getting started#

What jdx is, how to install it, your first queries, and wiring it into your AI agent.

• Introductionjdx is an IDE for AI agents, as a command line tool. It answers the questions an IDE answers for a human — what members a class has, what a method does, who…
• Installationjdx installs per user with one command, needs only a Java 21+ runtime, and runs on Linux, macOS and Windows.
• QuickstartFive minutes, two classpaths: the JDK (no setup at all) and a library pulled by its Maven coordinate. Every output below is real — it was produced by…
• Use with AI agentsWire jdx into Claude Code, Cursor, Codex, GitHub Copilot, Antigravity, OpenCode, Kilo Code, Cline or any MCP client with one command — or let any…

2Guides#

Task-oriented guides: naming symbols, choosing a classpath, reading and navigating code, upgrades, output, and serving.

• Symbol referencesEvery jdx command takes a symbol reference in Javadoc style. jdx is generous about what it accepts and canonical about what it prints, so any reference it…
• Classpath & workspacesjdx answers from a classpath made of jars, class directories, source directories, Maven coordinates and the JDK. This guide covers every way to supply one —…
• Reading codeLook at a type, list what you can call on it, read one method, or search the whole classpath — each with one command that returns only what was asked. This…
• Usages, hierarchy & callsFind who uses a symbol, what implements a type, who calls a method and what it calls, and real call sites to learn from — across every jar on the classpath,…
• API diff for upgradesBefore bumping a dependency, ask what the new version breaks. jdx diff compares the public API of two artifacts — jars, class directories or Maven…
• Output, JSON & exit codesjdx output is designed to be read by a language model with a finite context window: compact text by default, a complete JSON envelope on request, and exit…
• MCP, daemon, HTTP & batchThe one-shot CLI starts in about a quarter of a second. When an agent asks dozens of questions, keep jdx warm instead: an MCP server, a background daemon, a…

3Reference#

Every command and flag, generated from the jdx binary itself on every build.

• Command catalogueEvery jdx command with its IDE equivalent and key flags: read, search, graph, API diff, workspaces, serving.
• Agent cheat sheetThe compact jdx block to paste into CLAUDE.md, AGENTS.md or a system prompt: every command, the ref syntax and the exit codes.
• MCP toolsEvery tool the jdx MCP server exposes to Claude Code, Cursor, Codex and other MCP clients, with parameters — generated from jdx mcp.

Commands: version upgrade doctor show members outline body source signature doc search resolve ls tree usages hierarchy implementors callers calls samples diff batch bench serve setup mcp help daemon ws cache kotlin

4Advanced#

The full design specification, the testing strategy, and how this agent-built project is run.

• Design specificationThe full jdx design: symbol grammar, command surface, output design, index, decompilation, Kotlin support, serving, security.
• Testing strategyHow jdx is tested: TDD core, javap-differential, property-based, metamorphic, fault-injection and corpus soak tests, mutation gates.
• Built by agentsjdx is a tool for AI agents that is itself written almost entirely by AI agents, working under rules a human owner sets. This page describes how that works…
• AGENTS.mdThe rules every contributor — human or AI agent — follows in the jdx repository: principles, locked decisions, exit codes, build and test gates.
• ContributingHow to contribute to jdx: conventions, code style, the command checklist, testing and documentation discipline.