Java JAR Decompiler & Reverse Engineering MCP Server
Enables decompilation and reverse engineering of Java JAR files through multi-engine decompilation, candidate evaluation, Maven project structuring, AST-based deobfuscation, and bytecode comparison.
README
Java JAR Decompiler & Reverse Engineering MCP Server
A production-ready Model Context Protocol (MCP) server written in Node.js for Java .jar package decompilation, multi-engine candidate benchmark evaluation, Maven project structuring, non-breaking syntax repair, GumTree Spoon AST analysis & obfuscation renaming, and ASM bytecode parity metrics evaluation.
📑 Table of Contents
- Pipeline Architecture & Reverse Engineering Workflow
- Candidate Evaluation & Scoring Model
- Directory Structure
- Setup & Installation
- MCP Client Configurations
- Complete MCP Tools Reference (9 Tools)
- Complete MCP Prompts Reference (5 Prompts)
- Execution Verification
1. Pipeline Architecture & Reverse Engineering Workflow
For every target Java .jar package, the MCP server executes a systematic 6-phase pipeline:
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ Phase 1: │ │ Phase 2: │ │ Phase 3: │ │ Phase 4: │ │ Phase 5: │ │ Phase 6: │
│ Multi-Engine │───>│ Quantitative │───>│ Mavenization │───>│ Non-Breaking │───>│ AST De- │───>│ ASM Bytecode │
│ Decompilation │ │ Evaluation │ │ & Structure │ │ Syntax Repair │ │ Obfuscation │ │ Comparison │
└─────────────────┘ └─────────────────┘ └─────────────────┘ └─────────────────┘ └─────────────────┘ └─────────────────┘
- Phase 1: Multi-Engine Decompilation (
decompile_jar)- Decompile target JAR using multiple engine variants (CFR 0.152, Vineflower 1.12.0, Fernflower, Procyon, JADX) into isolated output directories.
- Phase 2: Comparative Candidate Evaluation (
evaluate_and_mavenize_sources)- Test-compile each candidate output using
javac -g -parameters -proc:none -encoding UTF-8. - Calculate quantitative quality score based on Java source count, compiler error count, remaining
.classfiles, and inline warning comments.
- Test-compile each candidate output using
- Phase 3: Maven Project Structuring (
evaluate_and_mavenize_sources)- Copy winning candidate sources to
src/main/javaand resources tosrc/main/resources. - Generate production-ready
pom.xmlwith source/target compliance and compiler args (-g -parameters).
- Copy winning candidate sources to
- Phase 4: Non-Breaking Syntax Repair (
compile_maven_project&fix_compilation_errors_prompt)- Fix compilation errors reported in logs without modifying business logic (type casting, generic inference, sneaky throw casts, synthetic package-info artifacts).
- Phase 5: AST Obfuscation Detection & Refactoring (
run_ast_deobfuscation_pipeline)- Parse Abstract Syntax Tree using GumTree Spoon AST Diff (
gumtree-spoon-ast-diff-1.124.jar). - Detect obfuscated/synthetic variable and method identifiers (
var0,var1,arg0, closure captures, single-letter variables). - Apply context-aware, domain-accurate renames across files without altering functional behavior.
- Re-compile and re-scan AST to confirm 0 obfuscations remain.
- Parse Abstract Syntax Tree using GumTree Spoon AST Diff (
- Phase 6: ASM Bytecode Parity & Metrics Analysis (
compare_bytecode_and_analyze)- Perform ASM bytecode comparison against original JAR.
- Record metrics: File/Bytecode Match %, Business Context Similarity %, and Code Readability Score in log reports.
2. Candidate Evaluation & Scoring Model
The evaluate_and_mavenize_sources tool selects the optimal decompiled candidate engine using a quantitative multi-factor scoring formula:
$$\text{Score} = (\text{JavaFiles} \times 100) - (\text{JavacErrors} \times 200) - (\text{RemainingClassFiles} \times 50) - (\text{DecompilerWarnings} \times 10)$$
Detailed Evaluation Criteria Breakdown:
- Compilation Pass Rate under Debug Flags (Weight: -200 per error):
- Each candidate is test-compiled using:
javac -g -parameters -proc:none -encoding UTF-8 - Rationale: Penalizes engines that emit invalid syntax, generic type inference failures, or broken lambdas. Zero compilation errors provides a massive score advantage.
- Each candidate is test-compiled using:
- AST Source Coverage (Weight: +100 per
.javafile):- Counts total
.javasource files successfully reconstructed. - Rationale: Rewards decompilers that fully reconstruct class structures, inner classes, and interface hierarchies without dropping files.
- Counts total
- Unhandled Bytecode / Remaining
.classFiles (Weight: -50 per.classfile):- Counts
.classbinary files left behind in output directory. - Rationale: Severe penalty for decompiler engines that fail on complex bytecode constructs (e.g., Kotlin synthetic bridges, inner classes) and leave un-decompiled
.classfiles.
- Counts
- Inline Decompiler Warning Comments (Weight: -10 per warning):
- Scans top 150 lines of every
.javafile for issue markers:// FAILED to decompile method ...// Could not decompile ...// Exception decompiling .../* Synthetic */
- Rationale: Penalizes engines that emit partial method stubs or swallow exceptions inline.
- Scans top 150 lines of every
3. Directory Structure
server/
├── bin/
│ └── cli.js # Executable CLI entry point for NPX / global execution
├── src/
│ ├── index.js # Main MCP server initialization & handler registration
│ ├── decompilerHandler.js # Core decompiler execution, Maven build, AST, & ASM engine
│ ├── tools/
│ │ └── definitions.js # Structured MCP Tool declarations & JSON schemas (9 tools)
│ └── prompts/
│ └── definitions.js # Structured MCP Prompt declarations & templates (5 prompts)
├── scripts/ # Utility helper scripts
├── index.js # Forwarding entrypoint (imports src/index.js)
├── package.json # Package manifest & configuration
└── README.md # Server documentation
4. Setup & Installation
Prerequisites
- Node.js:
v18.0.0or higher - Java Development Kit (JDK): JDK 8 / JDK 17 / JDK 24 (configured in
JAVA_HOMEor path) - Apache Maven:
mvnCLI installed and available in environment path
Installation
cd server
npm install
5. MCP Client Configurations
Claude Desktop Configuration
Add the server entry to your claude_desktop_config.json:
{
"mcpServers": {
"jar-decompiler": {
"command": "node",
"args": [
"C:/Users/ghosh/OneDrive/Desktop/Decompilation/server/index.js"
]
}
}
}
Antigravity CLI / Gemini Config
Add the server entry to your .gemini/antigravity-cli/mcp_config.json:
{
"mcpServers": {
"jar-decompiler": {
"command": "node",
"args": [
"C:/Users/ghosh/OneDrive/Desktop/Decompilation/server/index.js"
]
}
}
}
6. Complete MCP Tools Reference (9 Tools)
1. decompile_jar
Decompiles a Java .jar file into a target directory and returns comprehensive decompilation analytics, tree output, and logs.
- Parameters:
jarPath(string, required): Path to.jarfile.outputDir(string, optional): Destination directory.decompilerPath(string, optional): Explicit path to decompiler.jar.decompilerType(string, optional):'auto','cfr','vineflower','fernflower','procyon','jadx', or'generic'.extraArgs(array of strings, optional): Additional command-line flags.
2. list_decompilers
Lists all Java decompiler files (.jar or executables) currently found in the decompiler/ folder.
- Parameters:
decompilerDir(string, optional): Path to decompiler folder.
3. analyze_decompilation_output
Analyzes an existing directory containing decompiled source code to count Java files, detect decompilation warnings/errors, and produce a directory tree preview.
- Parameters:
outputDir(string, required): Path to decompiled directory.
4. evaluate_and_mavenize_sources
Evaluates decompiled outputs (comparing AST structure, line count, compiler warning count), chooses the optimal candidate using the quantitative scoring model, and structures it into a clean Maven project with pom.xml.
- Parameters:
outputsDir(string, required): Path to directory containing candidate folders (e.g.outputs/).targetMavenDir(string, required): Path to target directory (e.g.mavenized_merged_source/).groupId(string, optional): Maven groupId (default:org.apache.commons).artifactId(string, optional): Maven artifactId (default:commons-io).version(string, optional): Maven version (default:2.22.0).
5. compile_maven_project
Compiles a Maven project using mvn clean compile, parses compilation errors into a human-readable format, and writes the log file.
- Parameters:
projectDir(string, required): Path to Maven project root.logPath(string, optional): Path for compilation error log file.
6. compare_bytecode_and_analyze
Performs ASM bytecode analysis comparing the original JAR against the recompiled mavenized source, outputting percentage match, business context similarity, and variable readability scores.
- Parameters:
originalJarPath(string, optional): Path to original JAR.mavenDir(string, optional): Path to mavenized project root.logPath(string, optional): Path to output comparison report.asmJarPath(string, optional): Path to ASM library JAR.
7. generate_ast_and_detect_obfuscation
Uses GumTree Spoon AST Diff to parse Java source files, generate AST representations, and detect obfuscated or synthetic variable/method names (var0, arg1, single-letter variables, closure captures).
- Parameters:
sourceDir(string, optional): Path to Java source directory.gumtreeJarPath(string, optional): Path to GumTree Spoon JAR.logPath(string, optional): Path to output obfuscation report.
8. rename_obfuscated_variables
Copies mavenized source to final output directory, applies obfuscated variable renames with meaningful names, adds changelog comments to modified files, and generates a comprehensive rename log. NEVER modifies business logic.
- Parameters:
renames(array of objects, required): Array of{ file, oldName, newName, line }.sourceDir(string, optional): Source directory.targetDir(string, optional): Target output directory.logPath(string, optional): Output log file path.
9. run_ast_deobfuscation_pipeline
Runs the complete end-to-end AST de-obfuscation pipeline: copies mavenized_merged_source to mavenized_final_output, compiles, parses AST via GumTree Spoon to find obfuscated vars/methods, applies context-aware renames without changing business logic, verifies compilation, re-scans AST, and outputs logs to logs/ast_renamed_variables_methods.txt.
- Parameters:
sourceDir(string, optional): Source directory.targetDir(string, optional): Target output directory.gumtreeJarPath(string, optional): GumTree Spoon JAR path.logPath(string, optional): Output report log file path.renames(array of objects, optional): Custom rename entries array.
7. Complete MCP Prompts Reference (5 Prompts)
1. evaluate_and_mavenize_prompt
System prompt for comparing decompiled AST outputs, selecting the optimal candidate with minimal code loss, and mavenizing it.
- Arguments:
outputsDir,targetMavenDir
2. fix_compilation_errors_prompt
Mandatory system prompt for fixing Java compilation errors without modifying business logic (syntax, generic, and type-cast fixes only).
- Arguments:
logFile,projectDir
3. compare_bytecode_prompt
System prompt for running ASM bytecode comparisons between the original JAR and recompiled mavenized project, calculating similarity metrics and debug readability scores.
- Arguments:
originalJarPath,mavenDir,logPath
4. rename_obfuscated_variables_prompt
System prompt for AST-based detection and renaming of obfuscated/synthetic variable names in decompiled Java source, preserving all business logic.
- Arguments:
sourceDir,targetDir,logPath
5. ast_deobfuscation_pipeline_prompt
Comprehensive system prompt for copying source to final output, compiling baseline, building GumTree Spoon AST, detecting obfuscated vars/methods, applying context-aware renames, verifying build success, and writing log files.
- Arguments:
sourceDir,targetDir,logPath
8. Execution Verification
Run syntax checks across all server modules:
node --check index.js; node --check src/index.js; node --check src/decompilerHandler.js; node --check src/tools/definitions.js; node --check src/prompts/definitions.js; node --check bin/cli.js
Expected Output: Exit code 0 (clean pass).
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。