claude-ai-music-skills: a Claude Code plugin that turns a chat into an album pipeline
Human + AI music production workflow for Suno - skills, templates, and tools
At a glance
- What is it?
- The bitwize-music-studio project packages 53 Claude Code skills, an 80+ tool MCP server and a Suno-oriented mastering chain into one installable plugin. It is a serious piece of plugin engineering wrapped around a workflow that stops the moment you need to generate audio.
- Who is it for?
- Adopt it if you already write lyrics and generate on Suno and want the research, structure and mastering steps scripted inside Claude Code, and if you are on a Max subscription or ready to pace yourself around rate limits. Skip it if you want a tool that makes audio for you, if you refuse to install a Python 3.11+ venv and ffmpeg, or if you will not accept the human sign-off gate the research pipeline enforces.
- Can I use it commercially?
- Yes. CC0-1.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 3 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap between a lyric idea and a released track
The project's own framing is a person who loves music but never learned an instrument. That is the target user: someone who can describe an album concept, cannot play or mix it, and is willing to work in a terminal. The plugin does not generate audio. The README is explicit that you generate on Suno and everything else happens in the terminal, so the product is the surrounding pipeline: concept development, lyrics, Suno prompts, mastering, and release preparation.
The concrete problem is that those surrounding steps are where hobbyist AI music falls apart. Lyrics get written without checking which syllable the melody will land on. A Suno style prompt is guessed rather than assembled from genre tags and vocal direction. Stems come back from the generator and nobody measures loudness before uploading. The plugin answers each of those with a named stage and a checklist, which is a different proposition from a prompt pack or a list of genre tips.
Skills, model tiers and an MCP server as the nervous system
The architecture has three layers. The first is 53 skills, each a self-contained markdown file with YAML frontmatter declaring its model, description and activation conditions. Claude routes to them by context, or you call them as /bitwize-music:<name>. The second layer is model assignment: 7 skills run on Opus, 30 on Sonnet, 16 on Haiku, on the stated rationale that lyric writing and Suno prompts define output quality while imports and clipboard work do not.
The third layer is a Python MCP server exposing 80+ tools. The README describes it as the plugin's nervous system and says skills call MCP tools instead of reading files directly, which keeps responses fast and state consistent. Tool groups cover state lookups and cache rebuild, lyrics analysis (syllable counting, readability, rhyme detection, cross-track repetition), audio processing, and a PostgreSQL-backed database for promo content.
That design has a visible cost. A skill that queries a server is not a skill you can read and copy. If the MCP server is not running, the plugin's core functionality is not available, which the requirements file makes plain by labelling the mcp and pyyaml pins as required for plugin core functionality.
Installing from the plugin marketplace and running setup
Installation goes through Claude Code's plugin marketplace rather than pip. The README gives these two commands, and the second references the marketplace name bitwize-music and the plugin of the same name.
/plugin marketplace add bitwize-music-studio/claude-ai-music-skills
/plugin install bitwize-music@bitwize-musicAfter that, the README says to run /bitwize-music:setup, which detects your environment and installs dependencies, then /bitwize-music:configure to set your artist name and workspace paths. Expect the setup step to be where platform problems surface, since the plugin's audio path depends on external binaries and on Python 3.11+ for the MCP server.
For a development checkout rather than a plugin install, the repository ships a Makefile whose default target prints usage. The venv target creates .venv and installs both requirement files, and make test runs pytest with coverage, with a local coverage floor of 75 that the Makefile comments describe as deliberately lower than CI's 80 because a single OS cannot execute the win32 and darwin branches.
make venv
make test
make lintThe requirements file also documents a single venv for all plugin features under ~/.bitwize-music/venv, and notes that the document-hunter skill needs a separate Playwright chromium install after pip. The file pins scipy and numpy at versions it says must not be bumped while Python 3.11 is supported.
Where the pipeline actually blocks you
The research system is the most opinionated part of the project and the most likely to frustrate. For documentary or true-story albums a lead researcher dispatches parallel sub-agents across 10 domain specializations, each trained on where to find primary sources, and a verification agent cross-checks claims before human review. The README states the full pipeline as: gather sources, verify citations, require human sign-off, then and only then allow lyrics generation. That is a hard gate, not a suggestion. If you want to write a song about a historical event from memory, the workflow is designed to stop you.
The second constraint is cost and rate limits. The README says the project works best on the Max subscription and that a standard Pro subscription will hit rate limits during multi-track sessions. Multi-agent research across model tiers is expensive in a way a single chat is not.
The third is the audio toolchain. Mastering, mixing and promo video depend on ffmpeg and on pinned Python packages, and the README notes that promo video tests mock ffmpeg and are verified by hand rather than continuously guarded. Sheet music PDF export is CI-verified on windows-latest, but AnthemScore transcription runs against a licensed install, and the README says the free trial exposes no CLI on any OS. If transcription is the feature you came for, that is a paid dependency the plugin cannot work around.
How it compares with a plain Suno prompt workflow
The honest alternative is not another plugin. It is a notes file and the Suno web interface. That approach has real advantages: nothing to install, no venv, no rate limit budget spent on coordination, and no gate between an idea and a draft. For a single track, it is faster.
The difference in approach appears at album scale. A notes file cannot count syllables against a prosody rule, cannot detect a repeated phrase across tracks, cannot measure loudness to a target, and cannot trace a lyric line back to a captured source. The plugin's 72 genre directories with production guides, mastering presets and reference artists are also something a personal notes file accumulates slowly and inconsistently.
A closer comparison is a DAW-based workflow with a dedicated mastering plugin. That gives you far more control over the final audio, but it assumes you can operate the DAW. This project's bet is that the bottleneck is judgement and structure, not knob turning, and it hands the actual audio generation to Suno. If your bottleneck is the mix, this is the wrong tool.
Maintenance, licence and what the release history says
The repository is not archived, and the last push was on 2026-09-12. The README states that active development happens on the develop branch and that main only receives tested, stable releases, so a clone of main is a stable snapshot rather than the current edge. Recent releases listed are v0.101.0 and v0.100.0 on 2026-07-21 and v0.99.0 on 2026-07-13, a cadence of frequent minor bumps rather than long-lived versions.
Upgrade cost is dominated by the pinned dependency set. The requirements file holds scipy and numpy below versions it says require Python 3.12, with an explicit instruction not to bump past those lines while 3.11 is supported, and a reference to dependabot ignores enforcing that. Moving to Python 3.12 means revisiting those pins together, not one at a time. The Makefile's split between a local coverage floor of 75 and CI's 80 is a second thing to know before you file a coverage complaint.
The licence is CC0-1.0, which places the work in the public domain to the extent the licensor can do so. That is unusually permissive for a project of this size and removes the attribution question for derivative genre guides or templates. It does not affect the licences of the Python packages the plugin installs, which carry their own terms, nor the terms of Suno or AnthemScore. This is a description of what the repository states, not legal advice.
Editorial conclusion
Adopt it if you already write lyrics and generate on Suno and want the research, structure and mastering steps scripted inside Claude Code, and if you are on a Max subscription or ready to pace yourself around rate limits. Skip it if you want a tool that makes audio for you, if you refuse to install a Python 3.11+ venv and ffmpeg, or if you will not accept the human sign-off gate the research pipeline enforces. Verify first that your platform is covered in reference/cross-platform/tool-compatibility-matrix.md, and check whether the AnthemScore transcription step is usable for you, since the README states its free trial exposes no CLI on any OS.
Frequently asked questions
Can claude.ai make music?
Not audio. The README describes the plugin as handling concept development, lyrics, Suno prompts, audio mastering and release prep, while you generate the music on Suno. It is a production pipeline around a generator, not a generator.
Is Claude Skills an AI agent?
In this project the skills are markdown files with YAML frontmatter that declare a model and activation conditions, and Claude routes to them by context. The research system does dispatch parallel sub-agents across 10 domain specializations, so agentic behaviour is part of the design rather than the whole of it.
Is there an AI tool that can help me create music?
Yes, in the sense that this plugin covers the steps around generation: 72 genre directories with production guides and mastering presets, lyric prosody and rhyme analysis, a 7-point audio QC, and Suno V5 style prompt construction. The audio itself still comes from Suno.
Is making music with AI legal?
The README does not make a legal claim. The plugin does include a pre-generation quality gate that requires sources verified and artist names cleared, and its research pipeline requires human sign-off before lyrics generation.
Community notes