daisyui-mcp

A typed, local Model Context Protocol server for official DaisyUI 5 documentation and agent skills. It is distributed as a normal Python wheel, works with Python 3.10+, and keeps a bundled offline snapshot for predictable development.

Published package: daisyui-mcp on PyPI

Install with uv

Install the MCP server as a user tool:

uv tool install daisyui-mcp

For optional skill-management dependencies:

uv tool install "daisyui-mcp[skills]"

Install with pip

Install the MCP server into the active Python environment:

python -m pip install daisyui-mcp

For optional skill-management dependencies:

python -m pip install "daisyui-mcp[skills]"

For local development, sync the committed lockfile:

uv sync --locked --extra dev --extra skills

Run

The default command serves MCP over stdio:

daisyui-mcp

To configure an MCP client, use the installed executable as its command:

/absolute/path/to/daisyui-mcp/.venv/bin/daisyui-mcp

The compatibility module mcp_server.py remains available for older checkout-based configurations.

MCP surface

Tools:

• list_components and get_component(name) preserve the original API.
• search_daisyui(query) searches component and skill metadata.
• list_skills() and get_skill(name) expose the bundled official skill.
• refresh_daisyui() refreshes current component docs with cached fallback.
• install_daisyui_skills(target) installs the official skill tree.
• daisyui_status() reports content origin, counts, and refresh time.

Resources:

• daisyui://components
• daisyui://skills
• daisyui://install
• daisyui://status

The server exposes the complete surface through one stdio MCP registration. No separate registration is required for individual tools or resources.

API examples

The server speaks MCP over stdio. After registering daisyui-mcp, an MCP client can call any of the following tools through that one connection. The second column shows the JSON arguments for each tool call:

Tool	Arguments	Demonstration
list_components	{}	Discover the available component names and summaries.
get_component	{"name": "button"}	Read the complete Markdown documentation for one component.
search_daisyui	{"query": "modal"}	Search component and official skill metadata.
list_skills	{}	List the bundled official DaisyUI skills.
get_skill	{"name": "daisyui"}	Read one official skill document.
refresh_daisyui	{}	Refresh official component content with cached fallback.
install_daisyui_skills	{"target": "./.agents/skills/daisyui"}	Install the official skill tree into a project-local target. Omit target to use the default.
daisyui_status	{}	Report content origin, counts, and refresh time.

For example, a client can call search_daisyui with:

{"query": "modal"}

Then it can call get_component with the selected component name:

{"name": "modal"}

The registered resources can be read directly by URI:

• daisyui://components — the component index.
• daisyui://skills — the official skill index.
• daisyui://install — DaisyUI installation guidance.
• daisyui://status — current content status.

The stable importable compatibility API is also available for Python integrations:

from daisyui_mcp.server import get_component, list_components

print(list_components())
print(get_component("button"))

Example requests to make from Codex after configuration include:

• “List the available DaisyUI components and show the button documentation.”
• “Search DaisyUI for modal-related guidance, then summarize the best match.”
• “Install the official DaisyUI skills into this project.”

The MCP server can install skills into the current project with an explicit tool call. The CLI equivalent is:

daisyui-mcp skills install
daisyui-mcp skills install --target ./custom/skills/daisyui

The default target is ./.agents/skills/daisyui. The installer fetches the official DaisyUI skill tree, validates archive paths, and falls back to the packaged SKILL.md if the network is unavailable.

Configure Codex

Install a reusable Codex profile template in the current project:

daisyui-mcp config install

The command writes daisyui-mcp.config.toml to the current directory. Copy it to $CODEX_HOME/daisyui-mcp.config.toml (usually ~/.codex/daisyui-mcp.config.toml), then run Codex from this project folder:

mkdir -p "${CODEX_HOME:-$HOME/.codex}"
cp daisyui-mcp.config.toml "${CODEX_HOME:-$HOME/.codex}/daisyui-mcp.config.toml"
codex --profile daisyui-mcp

The profile starts one stdio MCP server; its complete tools and resources are available after startup. As a one-time global alternative, register the same server directly:

codex mcp add daisyui-mcp -- daisyui-mcp serve

See examples/config.toml for the equivalent TOML.

DaisyUI source and refresh behavior

The runtime specification follows the official DaisyUI installation guide:

https://daisyui.com/docs/install/

Current component content is fetched from:

https://daisyui.com/llms.txt

daisyui-mcp refresh stores a JSON snapshot in the platform user cache. The server serves cached content after a successful prior refresh and bundled content when no cache exists or the network is unavailable. The package also bundles current component docs, including aura, megamenu, otp, and tooltip, plus the official skill entrypoint.

Development

uv sync --locked --extra dev --extra skills
uv run --extra dev ruff check .
uv run --extra dev ruff format --check .
uv run --extra dev mypy src tests
uv run --extra dev pytest
uv run --extra dev python -m build
uv run --extra dev twine check dist/*

The project uses a src layout, setuptools with setuptools-scm, strict MyPy, Ruff, pytest, and a committed uv.lock. Git tags such as v0.2.1 determine the package version used by the release workflow.

License

MIT. See LICENSE.