Recaf MCP
Enables LLMs to perform JVM bytecode analysis and editing by operating a Recaf workspace through MCP tools. Supports class browsing, decompilation, bytecode manipulation, search, cross-references, and refactoring operations.
README
<div align="center">
Recaf MCP
「"Turn Recaf into an API that LLMs can call"」
<br>
<br>
Recaf 4 MCP bridge — operate a Recaf workspace via MCP protocol for JVM bytecode analysis and editing.
<br>
A Java plugin runs inside Recaf exposing a local HTTP API; a Python MCP server turns that API into MCP tools consumable by LLM clients.
<br>
Tools · Quick Start · Bytecode Editing · Install
</div>
Tools
Workspace
get_workspace_info open_workspace close_workspace add_supporting_resource list_supporting_resources fetch_current_class get_selected_text
Classes
get_all_classes get_class_info get_class_source get_bytecode_of_class get_methods_of_class get_fields_of_class get_inner_classes get_annotations_of_class get_raw_class_bytes
Methods
get_method_by_name get_method_bytecode get_method_info
Search
search_classes_by_name search_members_by_name search_strings search_numbers search_instructions
Cross References
get_xrefs_to_class get_xrefs_to_method get_xrefs_to_field get_callers_of_method get_callees_of_method get_overrides_of_method
Refactor
rename_class rename_method rename_field rename_package rename_local_variable apply_mappings
Decompilers
list_decompilers set_active_decompiler decompile_class_with
Inheritance
get_superclasses get_interfaces get_direct_subclasses get_all_subclasses get_implementors
Resources & Export
get_all_file_names get_file_content get_manifest get_strings_from_resources export_workspace get_modified_classes revert_class
Bytecode Editing
Full bytecode read/write: instruction-level editing, JASM text assembly, access flag editing, method/field CRUD, try-catch blocks, class byte replacement, workspace export.
Instructions
list_method_instructions get_method_bytecode replace_instruction insert_instruction remove_instruction
Method Body
assemble_method replace_method_body
Access Flags
edit_class_access edit_method_access edit_field_access
Method / Field CRUD
add_method remove_method add_field remove_field
Other
set_try_catch_blocks replace_class_bytes save_workspace
Requirements
- Java 25
- Python 3.10+
recaf.jar
Quick Start
1. Provide recaf.jar
The build looks for Recaf in this order:
RECAF_JARenv varlibs/recaf.jar../recaf/recaf.jar
$env:RECAF_JAR="D:\deobf\recaf\recaf.jar"
2. Build the Plugin
.\gradlew.bat jar
Output: build/libs/recaf-mcp-plugin-0.1.0.jar
3. Install Plugin
Copy the jar to Recaf's plugin cache directory %APPDATA%/Recaf/plugins/ and launch Recaf.
On successful load:
Recaf MCP plugin listening on http://127.0.0.1:8750
4. Install Python Dependencies
pip install -r requirements.txt
5. Start the MCP Server
python recaf_mcp_server.py --http --port 8751
Or use the launcher script:
.\start_recaf_mcp_http.ps1
MCP Config
{
"mcpServers": {
"recaf-mcp": {
"url": "http://127.0.0.1:8751/mcp"
}
}
}
Ports
| Component | Address |
|---|---|
| Plugin HTTP | 127.0.0.1:8750 |
| MCP HTTP | 127.0.0.1:8751/mcp |
Health check:
Invoke-WebRequest http://127.0.0.1:8750/health
Repository Structure
.
├── src/ # Java plugin source
├── mcp-server/ # Python MCP server
├── libs/ # recaf.jar goes here
├── recaf_mcp_server.py # MCP server entry point
├── requirements.txt # Python dependencies
├── .mcp.json # Project MCP config
└── README.md
Security
Binding to all interfaces exposes the MCP server with no authentication:
python recaf_mcp_server.py --http --host 0.0.0.0
Use localhost by default, or place it behind a firewall / VPN.
<p align="center"> <em>LLM Client -> MCP Server -> Recaf Plugin -> Recaf Workspace</em> </p>
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。