@tailwind-app/skills

Connects your coding agent to Tailwind and teaches it how to schedule Pinterest Pins.

One command installs two things: the Tailwind MCP server config, and a skill playbook that covers the parts of the API an agent gets wrong on its own.

npx @tailwind-app/skills

Supported harnesses: Claude Code, Cursor, and Codex.

What it writes

The installer detects which harnesses you have, asks before touching anything, and prints every path it wrote. Nothing else is modified.

Harness	Skill playbook	MCP server config
Claude Code	~/.claude/skills/tailwind-pinterest-scheduling/SKILL.md	via claude mcp add --scope user
Cursor	~/.cursor/skills/tailwind-pinterest-scheduling/SKILL.md	merged into ~/.cursor/mcp.json
Codex	~/.agents/skills/tailwind-pinterest-scheduling/SKILL.md	via codex mcp add

For Claude Code and Codex the installer calls each tool's own CLI rather than editing config files directly, so their merge semantics are theirs, not ours. Cursor has no equivalent command, so its mcp.json is edited in place — reading the existing file, adding only the tailwind key under mcpServers, and leaving every other server and top-level key untouched. If that file contains invalid JSON, the installer reports it and stops rather than overwriting something it could not parse.

After installing, restart your agent. Skill files are picked up live by Claude Code and Cursor, but all three read MCP server config at startup, and Cursor needs a Reload Window.

Claude plugin

This repository is also a Claude plugin. Claude Code users can install it through /plugin instead of running the command above, and it uploads to claude.ai under Settings > Customize > Plugins. Either way it provides the same two things — the same SKILL.md and the same remote MCP server — so pick whichever you prefer; running both the plugin and the installer is harmless but redundant.

The plugin is this repository root: .claude-plugin/plugin.json is the manifest and .mcp.json is the server entry, which points at the same https://mcp.tailwind.ai endpoint the installer configures. There is no bundled executable — a plugin install adds a remote server and a Markdown file, nothing that runs locally.

Options

--yes                 Non-interactive: install into every detected harness
--harness <name>      Install into a specific harness (claude-code, cursor, codex).
                      Repeatable, or comma-separated. Skips detection and the prompt.
--dry-run             Print what would be written; change nothing
--help                Show help
--version             Show the installed version

--dry-run is the honest way to see exactly what would land before it lands.

Authentication

Connecting uses OAuth — your agent opens Tailwind in a browser and you approve access. No API key is copied into a config file. If you'd rather use an API key, or your client can't do OAuth, see the setup documentation.

No telemetry

This installer sends nothing anywhere. It has no runtime dependencies, and once running it makes no outbound requests of its own — it reads and writes local files and shells out to CLIs you already have. The playbook ships inside the published package, so nothing is fetched from Tailwind's servers at any point, including on --dry-run.

To be precise about the one network step involved: npx downloads the package from npm the first time, exactly as it would for any package. That is npm fetching the tarball, not the installer calling home — and it is why a pinned version is fixed once you have it.

That's deliberate. This tool writes instructions your coding agent will follow, so it should be something you can read in full and pin, not something that changes underneath you.

Auditing and pinning

Every published version is immutable. To pin one:

npx @tailwind-app/skills@0.0.2

To read the playbook before installing anything, without running the installer at all:

npm view @tailwind-app/skills dist.tarball

Or browse the exact file that gets installed at unpkg.com/@tailwind-app/skills/skills/tailwind-pinterest-scheduling/SKILL.md.

Releases are published from this repository's own CI with npm provenance, so each published tarball is cryptographically linked to the commit and workflow that built it.

Removing it

Delete the skill file for your harness from the table above, then remove the MCP server:

claude mcp remove tailwind --scope user   # Claude Code
codex mcp remove tailwind                 # Codex

For Cursor, delete the tailwind entry under mcpServers in ~/.cursor/mcp.json.

License

MIT