mcpField
D
dejared

dejared

Local
github.com/hqkh4nhDeveloper Tools7Updated 10w ago

About dejared

dejared-mcp is a Java archive analyzer for AI coding agents and developers investigating libraries without source code. It browses archive contents, searches bytecode, extracts metadata, and decompiles classes with CFR, Vineflower, or Procyon.

Key features

  1. 01Browse packages, classes, and resources
  2. 02Read embedded text resources
  3. 03Search class names and bytecode strings
  4. 04Extract class and package metadata
  5. 05Decompile classes with three engines

Use cases

  1. 01Inspect a third-party library without source files
  2. 02Find embedded configuration or manifest data
  3. 03Trace a stack trace into a dependency
  4. 04Examine an obfuscated plugin or mod

MCP support

It is an MCP server exposing nine tools for archive discovery, bytecode search, metadata extraction, and class decompilation. It communicates with MCP clients over stdio.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"hqkh4nh-dejared": {
"command": "npx",
"args": [
"-y",
"dejared-mcp"
]
}
}
}
Run commandnpx -y dejared-mcp

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npmdejared-mcp1.2.1

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Java Node.js npm version npm downloads License: MIT

dejared-mcp: Java decompiler MCP server

dejared-mcp is a Java decompiler and JAR analyzer that speaks the Model Context Protocol. It lets an AI coding agent (Claude Code, Codex, Cursor, VS Code) open a .jar, .war, .ear, or .aar, list what is inside, search its bytecode, and decompile any class back to Java source using CFR, Vineflower, or Procyon.

Claude Code Codex Install in VS Code Add to Cursor

Table of Contents

Overview

dejared-mcp is a Java-based MCP server distributed as an npm package. It provides nine tools organized into three categories: discovery, hunting, and deep analysis. AI assistants use these tools to navigate JAR file structures, search for classes and string literals in bytecode, and decompile .class files back to readable Java source code.

Reach for it whenever a dependency ships without sources: reading what a third-party library actually does, recovering an application.yml or MANIFEST.MF embedded in a JAR, following a stack trace into a library you have no source for, or inspecting an obfuscated plugin or mod. It replaces hand-running jar tf, unzip, javap, or a desktop decompiler.

Any ZIP-based Java archive works: .jar, .war, .ear, .aar, and Spring Boot fat-jars.

The npm package acts as a thin wrapper that downloads and caches the server JAR on first run, then spawns it via java -jar using stdio transport.

Prerequisites

  • Node.js 20 or later
  • Java 17 or later (JRE is sufficient)

Quick Start

The best install is the plugin. One command pair gives your agent both halves: the nine MCP tools, and the jar-analysis skill that tells it when to reach for them instead of shelling out to jar tf, unzip, or javap.

claude plugin marketplace add hqkh4nh/dejared-mcp
claude plugin install dejared@dejared-mcp-marketplace

Codex carries the same bundle. See Install as a plugin. Every other client takes the plain MCP config.

Install as a plugin (recommended)

The MCP server on its own gives an agent nine tools. It does not tell the agent when to use them, so most agents keep reaching for jar tf, unzip -p, and javap -p out of habit and never touch the tools you installed.

A plugin ships both parts together:

PartWhat it does
MCP serverThe nine tools listed under Features
jar-analysis skillRoutes JAR questions to those tools, picks the cheap tool before the expensive one, and knows which decompiler to retry with

Two tools support that bundle today. Everything else is MCP-only.

Claude Code
claude plugin marketplace add hqkh4nh/dejared-mcp
claude plugin install dejared@dejared-mcp-marketplace

Inside a session, the same commands work as /plugin marketplace add ... and /plugin install .... If the install summary says Run /reload-plugins to activate, run it.

Codex
codex plugin marketplace add hqkh4nh/dejared-mcp
codex plugin add dejared@dejared-mcp-marketplace

Start a new session afterwards so the bundled skill and server load. Check with codex mcp list. dejared should be listed as enabled.

Install as an MCP server

For clients without a plugin system. You get the tools, but not the skill that steers the agent toward them.

Most clients accept this standard config:

{
  "mcpServers": {
    "dejared": {
      "command": "npx",
      "args": ["-y", "dejared-mcp"]
    }
  }
}
From the command line

If your tool has a CLI, one line is enough:

claude mcp add dejared -- npx -y dejared-mcp     # Claude Code
codex mcp add dejared -- npx -y dejared-mcp      # Codex
amp mcp add dejared -- npx -y dejared-mcp        # Amp
gemini mcp add dejared npx -y dejared-mcp        # Gemini CLI
Where the config file lives

Drop the standard config into the file for your tool:

ToolLocation
Claude Desktop%APPDATA%\Claude\claude_desktop_config.json (Windows) · ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
Claude Code.mcp.json in the project root
Cursor.cursor/mcp.json (project) · ~/.cursor/mcp.json (global)
WindsurfMCP servers panel in settings, or mcpServers in your settings file
ClineMCP settings file, via the extension's MCP Servers panel
Antigravity Editor~/.gemini/antigravity/mcp_config.json
Kiro.kiro/settings/mcp.json
Copilot CLI~/.copilot/mcp-config.json, or run /mcp add
Gemini CLI~/.gemini/settings.json (global) · .gemini/settings.json (project)
GooseAdvanced settings → Extensions → Add custom extension, type STDIO
JetBrains IDEsSettings → Tools → AI Assistant → Model Context Protocol (MCP)
Qodo GenChat panel → Connect more tools → + Add new MCP

A few setups need more than the standard config. Expand those below.

Codex: TOML instead of JSON
codex mcp add dejared -- npx -y dejared-mcp
codex mcp list

The -- matters: everything after it is the server's launch command.

Or edit ~/.codex/config.toml (global) or .codex/config.toml (project):

[mcp_servers.dejared]
command = "npx"
args = ["-y", "dejared-mcp"]

# Only if Java is not on your PATH
[mcp_servers.dejared.env]
DEJARED_JAVA_PATH = "/path/to/java"
VS Code (GitHub Copilot): uses servers, not mcpServers

Create or edit .vscode/mcp.json:

{
  "servers": {
    "dejared": {
      "command": "npx",
      "args": ["-y", "dejared-mcp"]
    }
  }
}

After saving, click Start above the server entry, then use Agent mode in Copilot Chat.

opencode: command is an array

Edit ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "dejared": {
      "type": "local",
      "command": ["npx", "-y", "dejared-mcp"],
      "enabled": true
    }
  }
}
Pure Java (no Node.js required)

If you prefer to run the server JAR directly without Node.js:

  1. Download the latest JAR from GitHub Releases.

  2. Run it:

    java -jar dejared-mcp-<version>.jar
    
  3. Configure your MCP client to use the JAR directly instead of npx:

    {
      "mcpServers": {
        "dejared": {
          "command": "java",
          "args": ["-jar", "/path/to/dejared-mcp-<version>.jar"]
        }
      }
    }
    

Features

Discovery

Browse and read JAR contents.

ToolDescription
dejared_list_packagesList all packages with class counts
dejared_list_classesList classes in a specific package
dejared_list_resourcesList non-class resource files
dejared_read_resourceRead text resources (YAML, XML, properties, JSON, and others)
Hunting

Search inside JAR files.

ToolDescription
dejared_search_classSearch classes by name
dejared_search_stringSearch string literals in bytecode (URLs, SQL, error messages)
Deep Analysis

Inspect metadata and decompile classes.

ToolDescription
dejared_get_metadataExtract class metadata via ASM (fast, no decompilation)
dejared_dump_package_metadataBatch metadata extraction for entire packages
dejared_decompile_classDecompile .class files to Java source code

Decompiler Engines

dejared-mcp supports three decompiler engines. The engine can be specified per request via the dejared_decompile_class tool.

EngineDescription
CFR (default)Reliable general-purpose decompiler
VineflowerModern fork of FernFlower, handles newer Java features well
ProcyonAlternative engine, can handle some edge cases better

Custom Java Path

If Java is not in your system PATH, set the DEJARED_JAVA_PATH environment variable in your MCP config. This applies to all npx-based configurations:

{
  "mcpServers": {
    "dejared": {
      "command": "npx",
      "args": ["-y", "dejared-mcp"],
      "env": {
        "DEJARED_JAVA_PATH": "/path/to/java"
      }
    }
  }
}

Server Configuration

PropertyDefaultDescription
dejared.cache.max-size500Max entries in the decompilation LRU cache
dejared.security.max-resource-size5242880Max resource file size (bytes)
dejared.security.decompile-timeout-seconds30Timeout per decompilation

How It Works

The npm package is a thin Node.js wrapper. On first run it:

  1. Checks the platform cache directory for a cached JAR matching the current version.
    • Linux: $XDG_CACHE_HOME/dejared-mcp (defaults to ~/.cache/dejared-mcp)
    • macOS: ~/Library/Caches/dejared-mcp
    • Windows: %LOCALAPPDATA%\dejared-mcp
  2. Downloads the JAR from GitHub Releases if not cached.
  3. Spawns java -jar with stdio inherited for MCP transport.

The server communicates over stdio using the Model Context Protocol.

FAQ and Troubleshooting

Q: Java is installed but the server cannot find it.

Set the DEJARED_JAVA_PATH environment variable in your MCP configuration. See Custom Java Path.

Q: The server fails to start with a permission error.

Ensure that the cached JAR file is readable. The cache location depends on your platform. See How It Works for the cache directory paths.

Q: Decompilation times out or returns an error.

Some classes are difficult to decompile. Try a different engine by specifying vineflower or procyon in the dejared_decompile_class tool. The default timeout is 30 seconds and can be adjusted via dejared.security.decompile-timeout-seconds.

Q: Which Java version do I need?

Java 17 or later. A JRE is sufficient; you do not need a full JDK.

Q: Can I run the server without Node.js?

Yes. Download the JAR from GitHub Releases and run it directly with java -jar. See the "Pure Java" section under Install as an MCP server.

Contributing

Contributions are welcome. Please follow these guidelines:

  1. Fork the repository and create a feature branch from master.
  2. Ensure your changes compile and pass existing tests.
  3. Write clear commit messages describing the purpose of each change.
  4. Open a pull request against master with a description of what the change does and why.

Every manifest listed under Project Structure carries a version field, and CI fails the build if any of them drifts from package.json. Bump them together.

Project Structure
dejared-mcp/
├── bin/                    # Node.js CLI entry point
├── lib/                    # Node.js wrapper (JAR download and process management)
├── java-mcp/               # Java MCP server (Spring Boot, Gradle)
│   └── src/
├── skills/                 # Agent skills, shared by every plugin manifest below
├── .mcp.json               # MCP server definition, shared the same way
├── .claude-plugin/         # Claude Code plugin + marketplace manifests
├── .codex-plugin/          # Codex plugin manifest
├── .agents/plugins/        # Codex marketplace manifest
├── package.json            # npm package manifest
└── server.json             # MCP Registry manifest
Building from Source
cd java-mcp
./gradlew build
Reporting Issues

Open an issue on GitHub Issues with steps to reproduce the problem, your Java version, and your Node.js version.

Third-Party Licenses

This project uses the following open-source libraries:

LibraryLicense
Spring BootApache 2.0
Spring AIApache 2.0
ASMBSD 3-Clause
CFRMIT
VineflowerApache 2.0
ProcyonApache 2.0

License

This project is licensed under the MIT License.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.