waves
Enables Claude to analyze and repair audio files by viewing spectrograms, measuring noise, and applying fixes like healing hums or isolating speakers, all with undoable operations and without uploading audio.
README
Waves for Claude
Repair audio by describing what's wrong — and let Claude look at the spectrogram.
This is the Claude plugin for MirrorEdge AI: Waves, a desktop audio-repair editor. The app is the MCP server: the same executable you click also answers on stdio, so Claude drives the real editor, through the same code and the same undo history your mouse uses.

Three bright bars are a phone buzzing through an interview. Three boxes go round them — one moment in time, each box its own band of pitch — and healing erases what is inside and nothing else. The voice underneath is untouched.
There's a hum through this whole interview and someone's phone buzzes about four minutes in. Clean it up and tell me what you did.
Claude opens your recording, measures it, notches out the hum, finds the buzz on the spectrogram, draws a box around it, heals only that, and reports the noise floor before and after. Everything it does is undoable.
Install
You need Waves itself — the plugin drives it, and the app is the server.
- Install Waves (free to use): Microsoft Store · direct download
- Add the marketplace and install the plugin:
/plugin marketplace add MirrorEdgeAI/waves-claude
/plugin install waves@mirroredge
That's it. The plugin finds your installation automatically — Store installs, direct installs and unusual locations are all handled. If it can't find Waves, it tells you where to get it.
Prefer to wire it up by hand? Open Waves and go to Help ▸ Works with Claude for the exact command for your machine.
What Claude can do
Seventeen tools. The two that matter most are at the bottom.
| Tool | What it's for |
|---|---|
open_recording |
Open an audio file — or a video, whose sound is lifted out |
close_recording |
Put it down again |
look_at_recording |
Run eleven checks and list what's wrong, worst first, in plain English |
measure |
Loudness, true peak, noise floor — where things stand right now |
list_fixes |
All 26 repairs, and what every setting means |
apply_fix |
Run one, on the whole recording or one stretch of it |
edit_box |
Draw a box in time and pitch, and heal, quieten or boost only inside it |
isolate_speakers |
Split a noisy room into one track per voice |
history · undo · redo |
Everything done so far, and walking it backwards and forwards |
export · export_speakers · save_session |
Keep the result |
check_the_export |
Confirm what landed on disk is what you meant |
picture_of_the_sound |
Claude sees the spectrogram — the actual picture, zoomed where it asks |
picture_of_the_editor |
Claude sees the app — any pane, as it looks on screen |
Those last two are why this is different from wrapping a command-line tool. Claude doesn't guess where the noise is from a description. It looks at the picture, finds the bright bar, and puts a box round it.
Your audio never leaves your computer
Claude sends instructions — "take out the clicks", "quieten 2–3 kHz between 0:06 and 0:08" — and Waves does the work locally, sending back measurements and, when asked, a picture. Your recording is never uploaded, transmitted or copied anywhere.
Waves makes exactly two kinds of network call in its entire life: activating your licence once, and fetching a file you asked it to open from a link.
What it costs
Nothing extra for the plugin. Waves is free to use — opening, editing, listening, measuring, and letting Claude experiment as much as it likes. A subscription covers what you keep: exported files, saved speaker tracks and saved sessions.
- $29/month or $290/year, direct from mirroredgeai.com/waves
- Mac App Store pricing is set by Apple and differs
Requirements
- Windows 10 (1809) or later, or macOS
- Waves 0.5.1 or later
- Claude Code, or any MCP client
Built by MirrorEdge AI · Guide · Terms · Privacy
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。