← Atlas · Topics: components, data-viz, agent-skills
BoardUI
- Site
- boardui.com
- Type
- design system
- Verdict
- Very useful
- Licence
- Conditional licence, mixed
mixedThe base components and blocks are MIT and installable without an account; BoardUI Pro unlocks chart cards, agent components and full-page templates, with duration and updates depending on the plan at review - Agent channels
- mcp, llms-txt, skill, registry
- Pricing
- freemium
- Reviewed
What it is
A design system aimed at two shapes that are usually served by different libraries: the data-heavy dashboard and the agentic interface. It ships 103 components, nineteen of them charts, plus an app shell, tables and 400+ tokens designed in Figma, on React and Tailwind v4.
Components install as source. npx boardui init writes tokens, global styles and agent rules; npx boardui add button writes the .tsx files and installs the packages they need. After that BoardUI is not a dependency to upgrade, it is code you edit — which is the right contract for a design that is expected to diverge from the defaults.
The agent story is unusually complete for a component library: an MCP server over HTTP with read-only tools (list_components, search_components, get_component, get_usage_examples, get_theme), a stdio variant that can also write files and activate a Pro licence, an installable skill, llms.txt, llms-full.txt and an OpenAPI document. init also writes always-on design rules into AGENTS.md and .cursor/rules/, including semantic tokens only and no hardcoded dark-mode colours.
When to open it
Open it when the product is a dashboard, admin panel or internal tool and the charts, tables and shell have to agree, or when you are building a chat or agent surface and need the tedious parts — a composer with attachments, streaming task lists, thinking indicators, context meters, web-search trails.
Most useful
- Charts and shell designed together, which is where most dashboard work actually goes wrong.
- Agent components as a category, not a theme: the states an agent UI needs are components here.
- Source install over runtime package, so a component can be changed without a fork.
- Design rules written into the repository by
init, which is how a library keeps an agent honest after the first file. - Four agent channels, with a read-only HTTP server for planning and a stdio server that can install.
llms-full.txtwith every component inlined, which is the right shape for a long-context agent.
Using it with agents
The strongest set in the atlas at review: llms.txt as a site map, llms-full.txt as the whole catalogue, an OpenAPI document, an MCP server in two transports, and a skill that installs a SKILL.md with the catalog, theming and page patterns. An agent can therefore discover, plan and install without a human copying anything.
Watch out for
- The free tier is the base set; chart cards, agent components and templates are Pro, which is exactly the part the product's positioning sells.
- Pro licence duration and updates depend on the plan, so the cost is not a one-time number.
- Semantic-token rules only help if your project adopts them; mixed token systems produce the drift the rules exist to prevent.
- Tailwind v4 is a hard requirement, which excludes projects still on v3.
- It is one opinionated system for dashboards: pushing it into a marketing site will fight its density.
Reusable ideas
- Write design rules into the repository at init, not into a docs page nobody opens.
- Offer
llms.txt,llms-full.txtand OpenAPI together; they serve different context budgets. - Split the MCP server into read-only and installing transports, and say which does what.
- Treat the agent interface as its own component category rather than a restyled chat box.