Android Source Explorer MCP Server
Provides on-demand access to AOSP and Jetpack source code to help AI understand Android framework internals. It features a hybrid architecture using Tree-sitter and LSP for precise code extraction, cross-file navigation, and local source indexing.
README
Android Source Explorer MCP Server
MCP server for exploring AOSP internals and Jetpack libraries.
<p align="center"> <img src="project-logo.png" width="600" alt="Android Source Explorer Logo"> </p>
Purpose
When building Android applications, AI tools often rely on outdated training data or incomplete summaries of the Android Framework. This MCP server provides on-demand, precise access to the actual source code (AOSP and AndroidX), enabling the AI to understand complex framework internals like the Activity lifecycle, ViewModel restoration, or Compose internals directly from the truth.
Installation & Setup
Prerequisites
- Python 3.11+
uv(recommended) orpip- Git
Initial Setup
-
Clone the repository:
git clone <repo-url> android-source-explorer cd android-source-explorer -
Install dependencies and perform an initial sync:
# Sync API 36 (Android 16) and common AndroidX packages uv run android-source-explorer sync --api-level 36 --androidx "compose,lifecycle,activity" # (Optional) Download LSP servers for cross-file features uv run android-source-explorer sync --lsp -
Check sync status:
uv run android-source-explorer status
Startup
The project provides a unified CLI. To start the MCP server for an AI client, use the serve command:
uv run android-source-explorer serve
Configuration
Add the server to your MCP client (e.g., Claude Desktop, Cursor, Gemini CLI).
Basic Configuration
{
"mcpServers": {
"android-sources": {
"command": "uv",
"args": ["run", "--directory", "/path/to/android-source-explorer", "android-source-explorer", "serve"]
}
}
}
With LSP Features Enabled
{
"mcpServers": {
"android-sources": {
"command": "uv",
"args": ["run", "--directory", "/path/to/android-source-explorer", "android-source-explorer", "serve"],
"env": {
"ANDROID_SOURCE_LSP": "true"
}
}
}
}
How It Works
1. Hybrid Architecture (Tree-sitter + LSP)
The server uses a dual-engine approach to balance speed and intelligence:
- Tree-sitter (Surgical Engine): Provides near-instant (sub-10ms) AST parsing of single files. It precisely extracts method bodies (including Javadoc and annotations), class members, and inheritance hierarchies without the overhead of a full compiler.
- LSP (Global Engine): Optionally leverages the Eclipse JDT LS (Java) and Kotlin Language Server for cross-file navigation. This enables finding references across millions of lines of code and resolving types that span multiple libraries.
2. Local Sync Strategy
To ensure all lookups are instant and offline-ready, the server uses a sync command to pre-fetch sources into a local cache (~/.android-sources/):
- AOSP: Fetches framework sources directly from
android.googlesource.comusing git sparse-checkouts. - AndroidX: Downloads
-sources.jarfiles from the Google Maven repository (dl.google.com/dl/android/maven2/). - Hybrid Indexing: Prioritizes your local
$ANDROID_HOMEsources if available, supplemented by the downloaded cache.
Available Tools
| Tool | Engine | Description |
|---|---|---|
search_classes |
Index | Search for classes by glob pattern or substring. |
lookup_class |
FS | Retrieve the full source code for a specific class. |
lookup_method |
Tree-sitter | Extract a precise method body + its Javadoc/annotations. |
list_class_members |
Tree-sitter | List all method and field signatures in a class. |
get_class_hierarchy |
Tree-sitter | Get the inheritance chain (superclass + interfaces). |
search_in_source |
FS/Regex | Search for text/regex across the entire source tree. |
goto_definition* |
LSP | Resolve the cross-file definition of a symbol. |
find_references* |
LSP | Find all usages of a class/method across the whole tree. |
get_type_info* |
LSP | Get documentation and type info via hover data. |
*Requires ANDROID_SOURCE_LSP=true
License
Apache License 2.0
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。