io.github.Algiras/debugium

编码与调试

by algiras

由 AI 驱动的实时调试器,提供 MCP bridge,支持 Python、JS、TS、Rust、Java 与 C/C++。

什么是 io.github.Algiras/debugium

由 AI 驱动的实时调试器,提供 MCP bridge,支持 Python、JS、TS、Rust、Java 与 C/C++。

README

Debugium

A multi-language debugger with a real-time web UI and LLM integration via MCP.

Debug Python, JavaScript, TypeScript, C, C++, Rust, Java, Scala, and WebAssembly programs from your browser — with AI-driven analysis through the Model Context Protocol.

CI Docs

Debugium UI — paused at a breakpoint in target_python.py with variables, call stack, breakpoints, watch, timeline, and console panels


Features

Web UI

  • Real-time web UI — source viewer, breakpoints, variables (recursive expansion), call stack, console, timeline, watch expressions, findings — all updated live via WebSocket
  • Multi-tab source viewer — open multiple files, click stack frames to navigate
  • Variable search — filter variables by name with recursive expansion
  • Changed-variable highlighting — variables that changed since last stop shown in orange
  • Thread selector — switch between threads in multi-threaded programs
  • Panel collapse & resize — drag to resize, toggle to collapse; Slim / Standard / Full layout presets
  • Dark / light mode toggle
  • Auto-reconnect — UI reconnects after a dropped WebSocket, with visual status indicator
  • Keyboard shortcuts — F5 continue, F10 step over, F11 step in, Shift+F11 step out, Ctrl/Cmd+D dark mode
  • Button animations — in-flight spinner and completion flash for debug commands

Debugging

  • Multi-language — Python (debugpy), Node.js/TypeScript (js-debug), C/C++/Rust (lldb-dap), Java (java-debug), Scala (Metals), WebAssembly (lldb-dap), or any DAP adapter via dap.json
  • Comprehensive DAP coverage — 35+ DAP requests: breakpoints, stepping, goto, memory read/write, disassembly, and more
  • Breakpoints — conditional, logpoints, hit-count, function, data (watchpoints), exception, run-to-cursor (continue_until)
  • Multi-session — debug multiple programs simultaneously, with child session routing (js-debug)
  • Remote debugging — attach to debugpy, JDWP, or Node inspector running on another machine or container

LLM / MCP Integration

  • 64 MCP tools — the full debug session exposed to Claude or any LLM
  • Capability-gated tools — tools automatically shown/hidden based on adapter capabilities
  • Compound toolsget_debug_context (orient in one call), step_until, step_until_change, run_until_exception, explain_exception, get_call_tree, compare_snapshots, find_first_change
  • Execution timeline — every stop recorded with changed variables and stack summary
  • Watch expressions — evaluated automatically at every breakpoint, manageable by LLM or UI
  • Annotations & findings — pin notes to source lines, record conclusions visible in the UI
  • Session export/import — save and restore debugging knowledge across sessions

CLI Control

  • Full CLI — 13 subcommands to drive sessions from a second terminal without the web UI
  • Auto-discovery — port file at ~/.debugium/port, session logs in ~/.debugium/sessions/

Install

Claude Code Plugin (recommended)

code
/plugin marketplace add Algiras/debugium
/plugin install debugium@debugium

Then add to your project's .mcp.json (see MCP Tools below).

macOS / Linux binary

bash
curl -fsSL https://raw.githubusercontent.com/Algiras/debugium/main/install.sh | bash

From source

bash
# Prerequisites: Rust stable + wasm-pack
cargo install wasm-pack

# Build UI
wasm-pack build crates/debugium-ui --target web --out-dir pkg
cp crates/debugium-ui/pkg/cm_init.js         crates/debugium-ui/dist/pkg/
cp crates/debugium-ui/pkg/debugium_ui.js      crates/debugium-ui/dist/pkg/
cp crates/debugium-ui/pkg/debugium_ui_bg.wasm crates/debugium-ui/dist/pkg/

# Build & install server
cargo install --path crates/debugium-server

Usage

Debug a Python file

bash
debugium launch my_script.py --adapter python

Debug a Node.js / TypeScript file

bash
debugium launch app.js --adapter node
debugium launch app.ts --adapter typescript

Debug C / C++ / Rust

bash
# C or C++ (compile with -g for debug info)
cc -g -O0 main.c -o main && debugium launch ./main --adapter lldb
c++ -g -O0 main.cpp -o main && debugium launch ./main --adapter lldb

# Rust
cargo build && debugium launch target/debug/my_binary --adapter lldb

Debug Java / Scala

bash
# Java (requires microsoft/java-debug adapter)
debugium launch MainClass --adapter java

# Scala (connect to a running Metals DAP server)
debugium launch build-target --adapter metals
debugium launch build-target --adapter metals:5005  # custom port

Attach to a running process (remote debugging)

bash
# Python (debugpy listening on port 5678)
debugium attach --port 5678 --adapter python

# Java (JDWP on port 5005)
debugium attach --port 5005 --adapter java

# Node.js (inspector on port 9229)
debugium attach --port 9229 --adapter node

Or via MCP: attach_session(port=5678, adapter="python", breakpoints=["/path/app.py:42"])

Use a custom adapter via dap.json

bash
# Create a dap.json (see dap.json.example) then:
debugium launch my_program --config ./dap.json

# Or place dap.json in cwd / .debugium/ for auto-discovery:
debugium launch my_program   # finds ./dap.json automatically

Set initial breakpoints

bash
debugium launch my_script.py --adapter python \
  --breakpoint /abs/path/my_script.py:42 \
  --breakpoint /abs/path/helpers.py:15

Enable LLM / MCP integration

Add a .mcp.json to your project root (Claude Code picks this up automatically):

json
{
  "mcpServers": {
    "debugium": {
      "command": "debugium",
      "args": ["mcp"]
    }
  }
}

Then launch the session normally — the MCP server connects to whichever port is active:

bash
debugium launch my_script.py --adapter python --breakpoint /abs/path/my_script.py:42

Claude Code will now have access to all Debugium MCP tools. See CLAUDE.md for the recommended workflow and SKILL.md for the full tool reference.


CLI Control Commands

Once a session is running (debugium launch …), you can drive it from a second terminal — or from an LLM agent — without touching the web UI.

Port is auto-discovered from ~/.debugium/port; override with --port.

Global flags (all subcommands)

FlagDefaultDescription
--port PORT~/.debugium/portServer port to connect to
--session IDdefaultSession to target
--jsonoffPrint raw JSON instead of human-readable output

Inspection

bash
debugium sessions                  # list active sessions
debugium threads                   # list threads
debugium stack                     # show call stack
debugium vars                      # show local variables (auto-resolves top frame)
debugium vars --frame-id 2         # show variables for a specific frame
debugium eval "len(fibs)"          # evaluate expression in top frame
debugium eval "x + 1" --frame-id 2
debugium source path/to/file.py    # print full source file
debugium source path/to/file.py --line 43  # windowed ±10 lines with → marker
debugium context                   # full snapshot: paused-at, stack, locals, source, breakpoints
debugium context --compact         # same but truncated (3 frames, 10 vars)

Breakpoints

bash
debugium bp set FILE:LINE [FILE:LINE …]   # set breakpoints (replaces existing in that file)
debugium bp list                          # list all breakpoints
debugium bp clear                         # clear all breakpoints

Execution control

bash
debugium continue                  # resume execution
debugium step over                 # step over (next line)
debugium step in                   # step into a function call
debugium step out                  # step out of current function

UI annotations (visible in the web UI)

bash
debugium annotate FILE:LINE "message" [--color info|warning|error]
debugium finding "message"         [--level  info|warning|error]

Example workflow

bash
# Terminal A — start the session
debugium launch tests/target_python.py --adapter python \
  --breakpoint "$(pwd)/tests/target_python.py:43"

# Terminal B (or LLM agent) — inspect and drive it
debugium sessions
debugium stack
debugium vars
debugium eval "len(fibs)"
debugium bp set tests/target_python.py:49
debugium continue                  # runs to line 49
debugium vars
debugium step over
debugium context --json            # machine-readable snapshot
debugium annotate tests/target_python.py:43 "called here" --color info
debugium finding "fibs has 10 elements" --level info
debugium bp clear

MCP Tools

When connected via MCP, 64 tools are available. Key ones:

CategoryTools
Orientget_debug_context ★ (paused location + locals + stack + source in one call)
Breakpointsset_breakpoint, set_breakpoints, set_logpoint, list_breakpoints, clear_breakpoints, set_function_breakpoints, set_exception_breakpoints, set_data_breakpoint, list_data_breakpoints, clear_data_breakpoints, breakpoint_locations
Executioncontinue_execution, step_over, step_in, step_out, pause, goto, disconnect, terminate, restart
Inspectionget_stack_trace, get_scopes, get_variables, evaluate, get_threads, get_source, get_capabilities, loaded_sources, source_by_reference, step_in_targets
Mutationset_variable, set_expression
Outputget_console_output, wait_for_output (with from_line to avoid stale matches)
Memoryread_memory, write_memory, disassemble (native debugging)
Historyget_timeline, get_variable_history, compare_snapshots, find_first_change
Annotationsannotate, get_annotations, add_finding, get_findings
Watchesadd_watch, remove_watch, get_watches
Compoundstep_until, step_until_change, continue_until, run_until_exception, explain_exception, get_call_tree, restart_frame
Sessionget_sessions, list_sessions, launch_session, attach_session, stop_session, export_session, import_session
Controlgoto_targets, cancel_request

Note: step_over, step_in, and step_out are blocking — they wait for the adapter to pause before returning. Safe to chain back-to-back without sleeps. continue_execution returns console_line_count for use with wait_for_output. Tools like read_memory, goto, and restart_frame only appear when the adapter supports them.

See SKILL.md for the full reference with input schemas.


Keyboard Shortcuts

KeyAction
F5Continue
F10Step Over
F11Step Into
Shift+F11Step Out
Ctrl/⌘+DToggle dark/light mode

Architecture

code
debugium-server (Rust + Axum)
├── DAP proxy     — spawns / attaches to debug adapters (debugpy, js-debug, lldb-dap, java-debug, Metals, custom)
├── HTTP API      — /state, /sessions, /annotations, /findings, /watches, /timeline
├── WebSocket     — broadcasts DAP events + enriched stop data (changed vars, timeline) to UI
├── MCP stdio     — JSON-RPC 2.0 server exposing 64 tools for LLM integration
├── CLI control   — 13 subcommands to drive sessions from a second terminal
└── ~/.debugium/  — port file, session logs (events.ndjson), debug log

debugium-ui (Leptos + WASM)
├── CodeMirror 6  — source viewer with breakpoint gutters, exec arrow, LLM annotations, multi-tab
├── Reactive panels — Variables, Stack, Breakpoints, Findings, Watch, Timeline, Console (18 components)
└── WebSocket     — receives events, sends DAP commands, auto-reconnects with status indicator

Supported Languages & Adapters

Language--adapter flagPrerequisiteVerified
Pythonpython / debugpypip install debugpy
Node.jsnode / jsjs-debug (bundled or build from vscode-js-debug)
TypeScripttypescript / ts / tsxjs-debug + tsx or ts-node in PATH
C / C++lldb / codelldblldb-dap (Xcode on macOS; apt install lldb on Linux)
Rustlldb / rustlldb-dap + cargo build
Javajava / jvmmicrosoft/java-debug adapter JAR
Scalametals / scalaRunning Metals DAP server⚠️ (requires running Metals)
WebAssemblywasmlldb-dap (LLVM ≥16)⚠️ (requires WASM-aware LLVM)
Any DAP adapter--config dap.jsonSee dap.json.example

Remote debugging

Connect to a DAP server running on another machine (or in a container):

json
{
  "adapterId": "debugpy",
  "request": "attach",
  "host": "192.168.1.100",
  "port": 5678,
  "pathMappings": [{ "localRoot": ".", "remoteRoot": "/app" }]
}
bash
debugium launch app.py --config remote.json

License

MIT

常见问题

io.github.Algiras/debugium 是什么?

由 AI 驱动的实时调试器,提供 MCP bridge,支持 Python、JS、TS、Rust、Java 与 C/C++。

相关 Skills

网页构建器

by anthropics

Universal
热门

面向复杂 claude.ai HTML artifact 开发,快速初始化 React + Tailwind CSS + shadcn/ui 项目并打包为单文件 HTML,适合需要状态管理、路由或多组件交互的页面。

在 claude.ai 里做复杂网页 Artifact 很省心,多组件、状态和路由都能顺手搭起来,React、Tailwind 与 shadcn/ui 组合效率高、成品也更精致。

编码与调试
未扫描114.1k

前端设计

by anthropics

Universal
热门

面向组件、页面、海报和 Web 应用开发,按鲜明视觉方向生成可直接落地的前端代码与高质感 UI,适合做 landing page、Dashboard 或美化现有界面,避开千篇一律的 AI 审美。

想把页面做得既能上线又有设计感,就用前端设计:组件到整站都能产出,难得的是能避开千篇一律的 AI 味。

编码与调试
未扫描114.1k

网页应用测试

by anthropics

Universal
热门

用 Playwright 为本地 Web 应用编写自动化测试,支持启动开发服务器、校验前端交互、排查 UI 异常、抓取截图与浏览器日志,适合调试动态页面和回归验证。

借助 Playwright 一站式验证本地 Web 应用前端功能,调 UI 时还能同步查看日志和截图,定位问题更快。

编码与调试
未扫描114.1k

相关 MCP Server

GitHub

编辑精选

by GitHub

热门

GitHub 是 MCP 官方参考服务器,让 Claude 直接读写你的代码仓库和 Issues。

这个参考服务器解决了开发者想让 AI 安全访问 GitHub 数据的问题,适合需要自动化代码审查或 Issue 管理的团队。但注意它只是参考实现,生产环境得自己加固安全。

编码与调试
83.4k

by Context7

热门

Context7 是实时拉取最新文档和代码示例的智能助手,让你告别过时资料。

它能解决开发者查找文档时信息滞后的问题,特别适合快速上手新库或跟进更新。不过,依赖外部源可能导致偶尔的数据延迟,建议结合官方文档使用。

编码与调试
52.2k

by tldraw

热门

tldraw 是让 AI 助手直接在无限画布上绘图和协作的 MCP 服务器。

这解决了 AI 只能输出文本、无法视觉化协作的痛点——想象让 Claude 帮你画流程图或白板讨论。最适合需要快速原型设计或头脑风暴的开发者。不过,目前它只是个基础连接器,你得自己搭建画布应用才能发挥全部潜力。

编码与调试
46.3k

评论