Command Palette

Search for a command to run...

UnylyUnyly
Весь каталог

Solid

БесплатноПоддерживается

SolidJS components for IconMind — 5,287 tree-shakeable icons for AI-era software (LLMs, agents, MCP, RAG).

GitHubEmbed

Описание

SolidJS components for IconMind — 5,287 tree-shakeable icons for AI-era software (LLMs, agents, MCP, RAG).

README

156 IconMind icons taken evenly across the set: agents, models, vector databases, MCP servers, charts, files and arrows

IconMind

Open-source icons for AI-era software.
LLMs, agents, MCP, RAG, and everything around them.

Website · Browse all icons · Collections · Docs · MCP server

MIT license icons cells npm pub.dev


import { Agent, ContextWindow, VectorDatabase } from "@iconmind/react";

<Agent />
<ContextWindow variant="duotone" weight="bold" />
<VectorDatabase size={32} className="text-violet-500" />

5,287 icons · 31,722 drawings · 10 packages · one generated source of truth.

Every icon is a declaration compiled into six cells by a validator that refuses geometry it cannot draw correctly, and a nightly job rasterises all 31,722 of them and fails if any two icons render the same. That is how a set this size still reads as one hand, and why an assistant can be handed an MCP server that returns real names instead of <AgentBrain />.

Icons get redrawn when that scan or the contact sheet catches them — a whole round of rooms and appliances went back on the board because shapes were fusing at 16 px — and Bluetooth was dropped because its rune needs three stroke crossings where the set allows two. Everything else that was turned away is written down too.

Why this exists

Build a UI for an agent, a RAG pipeline, or an MCP server and you run out of icons almost immediately. There is no context-window, no reranker, no tool-calling, no mcp-resource in any mainstream set, so interfaces reach for a generic robot, a generic database, a generic lightning bolt, and the meaning is lost.

Lucide, Heroicons and Phosphor are excellent, and a thousand domain-specific AI icons would be scope creep for any of them. IconMind sits next to them, not instead of them. The 24px grid and 2px regular stroke match Lucide on purpose, so the two can share an interface without clashing.

IconMind Generalist sets
Vocabulary AI, agents, MCP, RAG, data, devops General UI
Grid 24px, 2px stroke Usually the same
Variants 6 drawings per concept Usually 1–2
Consistency Machine-enforced geometry Reviewed by eye
For AI assistants Built-in MCP server

The set

5,287 icons · 20 domains · 6 cells each. Two variants, outline and duotone, each drawn at three weights (1.5 / 2 / 2.5). A weight is a real drawing constraint, not a stroke-width slider: gap and legibility rules tighten as the stroke grows, and the validator checks every cell separately.

Duotone is derived, never drawn: every closed body tints at 20% behind the strokes, and icons made only of open marks (a check, an arrow) get a halo, the same paths echoed wider behind them, so the whole set answers the variant switch, not just the lucky icons.

Six icons drawn in all six cells: outline and duotone, each at thin, regular and bold

AI & LLM: models, prompts, tokens, embeddings, evaluation, safety Agents: planning, memory, tool use, computer use, multi-agent MCP: servers, clients, resources, tools, transports
RAG & Search: chunking, retrieval, ranking, vectors, grounding Data Engineering: pipelines, warehouses, streaming, quality DevOps: CI/CD, containers, observability, incidents
Cloud: compute, storage, network, edge, cost Security: auth, secrets, policy, AI security Automation: workflows, triggers, conditions, human-in-the-loop
Analytics: charts, metrics, experiments, LLM observability Developer Tools: code, terminal, VCS, API, testing Interface: arrows, actions, states, layout, media
Commerce: store, checkout, shipping, promotions, merchant Finance: banking, wallets, investing, accounting, insurance Travel: vehicles, transit, lodging, navigation, trips
Weather: sky, storms, temperature, nature, climate Health: medical, fitness, nutrition, mind, care Education: school, reading, courses, science, arts
Home: rooms, appliances, kitchen, garden, household Social: people, events, sport, games, hobbies

Install

Stack Install
React npm i @iconmind/react
Vue 3 npm i @iconmind/vue
Svelte npm i @iconmind/svelte
Solid npm i @iconmind/solid
Preact npm i @iconmind/preact
React Native npm i @iconmind/react-native react-native-svg
Astro npm i @iconmind/astro
Flutter flutter pub add iconmind_flutter
Laravel Blade composer require iconmind/blade-iconmind
Plain SVG / sprite npm i @iconmind/icons

One API, everywhere

Every framework package exposes the same component names and the same props, generated from one source so they cannot drift:

Prop Default What it does
size 24 Width and height
color currentColor Stroke colour
variant "outline" "outline" or "duotone"
weight "regular" "thin", "regular" or "bold", each its own drawing
strokeWidth per weight Fine adjustment; picking a weight redraws instead
absoluteStrokeWidth false Keeps stroke thickness constant as size changes

The same icon in four stacks:

// React / Preact / Solid — also Vue and Svelte with template syntax
<VectorDatabase variant="duotone" weight="bold" size={32} />
// Flutter — real strokes on a CustomPaint, not an icon font,
// so duotone and the weights survive intact
IconMind(IconMindIcons.vectorDatabase,
    variant: IconMindVariant.duotone, weight: IconMindWeight.bold, size: 32)
{{-- Laravel — server-rendered, zero JavaScript --}}
<x-im-vector-database-duotone-bold class="w-8 h-8" />
---
import { VectorDatabase } from "@iconmind/astro";
---
<VectorDatabase variant="duotone" weight="bold" size={32} />

Tree shaking, measured

Importing three icons costs ~660 bytes gzipped, not the whole library, and that is a test that fails in CI if it stops being true rather than a promise. Every icon is also its own entry point (@iconmind/react/icons/agent). In Flutter, icons are compile-time constants: the AOT compiler drops every one you never mention.

Download

Every icon page has a Download menu with SVG, PNG at 16 to 512 px, WebP, a favicon .ico, JPEG and Copy PNG, rendered in your browser from the exact variant, weight and colour on screen. The whole set ships as two archives on every GitHub Release: iconmind-svg.zip (all 31,722 cells) and iconmind-png.zip (one 512 px PNG per icon). It is also an Iconify collection (iconmind:agent) and one jsDelivr URL away with no install.

For AI assistants

There is an MCP server, so an assistant writing your UI can search the real set instead of guessing icon names that do not exist:

claude mcp add iconmind -- npx -y @iconmind/mcp

shadcn/ui. The set is also a shadcn registry item, so a shadcn project gets an <Icon name="agent-run" /> wrapper over @iconmind/react in one command, no approval from anyone:

npx shadcn@latest add https://iconmind.dev/r/iconmind.json

It runs offline, bundles the icon data, and starts in well under a second. The site also serves llms.txt and a full machine-readable inventory, so assistants without tool access can still pick real names.

No build step at all

@iconmind/icons ships every cell as a file on disk (icons/<category>/<slug>/<variant>-<weight>.svg), a sprite.svg of <symbol> elements, and the metadata database on its own subpath, so importing it is a decision rather than an accident:

<svg width="24" height="24"><use href="sprite.svg#im-agent" /></svg>
import { getIcon, iconsIn } from "@iconmind/icons/metadata"; // tools only: ~180 KB gz
import { version } from "@iconmind/icons";                    // a few hundred bytes

Design system

Icons here are written, not drawn: each one is a TypeScript declaration compiled into its six cells, and every rule a machine can hold is held by a machine. An illegal drawing cannot reach a file, because the shape constructors throw before it does:

24 × 24 canvas  ·  0.5 grid  ·  angles 0 / 45 / 90 only  ·  live area 2..22
round caps      ·  ≤ 2 stroke crossings  ·  ink centred within 2 units
longer side 18–22 across the whole set  ·  duotone tint 20%, halo = weight + 3

The vocabulary does the rest: a family shares one body and differs only in the mark inside it, so siblings are byte-identical where they agree.

Body Meaning Example
Open ring An agent, acting on its own agent, agent-check
Chamfered box A machine; the chamfer is MCP's mark model, mcp-server
Two-pronged plug A capability an agent can pick up tool-calling, mcp-tool
Page with a fold A document mcp-resource, policy
Squared shield Protection shield-check, encryption
Square loop Anything that goes round refresh, retry, sync
Cylinder Stored data database, backup

So you can usually guess the next icon in a family before you see it. The full reasoning, with the measurements behind each rule, is in the design guidelines.

Measured, not just validated

Beyond the validator, three tools keep the set honest at scale. All three grew out of real audits that redrew more than sixty icons:

pnpm icons:audit                  # size, ink density and centred-ness outliers, with named exceptions
pnpm icons:twins                  # renders all 5,287 icons and pixel-compares every pair for lookalikes
pnpm icons:duplicates --perceptual # the nightly scan: 31,722 cells, fails on two that render alike

Don't take my word for it

Every claim on this page is a command you can run after cloning:

pnpm icons:validate                # every rule, all 31,722 cells
pnpm icons:twins                   # pixel-compare every pair at 24px
pnpm icons:duplicates --perceptual # the nightly scan for lookalikes
pnpm test                          # including the bundle-size tripwires

Versioning & releases

All npm packages share one version (currently in lock-step with iconmind_flutter on pub.dev), and releases are fully automatic: every merged change steps the patch version, publishes all packages, tags, cuts a GitHub Release and redeploys the site. What you see on npm is never ahead of or behind the repo.

FAQ

Why is there no CommonJS build? Providing one would provide a path that does not tree-shake well, and some bundlers would silently choose it. ESM only.

Why only 0/45/90° angles? It is the constraint that makes 5,287 icons read as one hand. Some pictures are refused rather than drawn badly: a warning triangle at 45° is a tent, so warning is a circle. The rest of the refusals are listed here.

Can I request an icon? Yes, and it takes a minute and is the most useful contribution there is: request an icon.

Brand icons? No. Concepts age better than logos, and logos carry trademark weight this set does not want.

Is there a Figma library? No, and none is planned. It would be a second source of truth beside the compiler. The set is an Iconify collection (iconmind:agent), which Iconify's own Figma plugin reads, and every icon page copies or downloads its SVG.

How do I pick between IconMind and Lucide/Tabler/Heroicons? There are comparison pages with the counts, licences and trade-offs written out.

Contributing

Icons are TypeScript declarations in scripts/draw/icons/. The compiler and validator give you feedback in seconds, and a contact sheet renders any subset at 88/24/16px:

pnpm install
pnpm icons:build       # draw every icon from source
pnpm icons:validate    # every rule, all 31,722 cells
pnpm icons:audit       # the consistency measures

See the contributing guide for naming rules, originality policy and the pull-request checklist.

Ten icons are described down to their body and mark slot if you want one to start from. The drawing is the fun part; the rules are the tools' job.

Support the project

If this set saves you from mapping reranker to a generic sort icon, a ⭐ helps other people building AI interfaces find it.

License

MIT, code and icons alike. Free for commercial use, no attribution required, no conditions. Forever.

from github.com/iconmind/iconmind

Установить Solid в Claude Desktop, Claude Code, Cursor

Рекомендуется · одна команда, все IDE
unyly install solid

Ставит в Claude Desktop, Claude Code, Cursor и VS Code — сам разбирается с npx, uvx и сборкой из исходников.

Впервые? Поставь CLI: curl -fsSL https://unyly.org/install | sh

Или настроить вручную

Выполни в терминале:

claude mcp add solid -- npx -y @iconmind/solid

Пошаговые гайды: как установить Solid

FAQ

Solid MCP бесплатный?

Да, Solid MCP бесплатный — установка в пару кликов через Unyly без оплаты.

Нужен ли API-ключ для Solid?

Нет, Solid работает без API-ключей и переменных окружения.

Solid — hosted или self-hosted?

Self-hosted: сервер запускается локально на твоей машине командой из раздела установки.

Как установить Solid в Claude Desktop, Claude Code или Cursor?

Открой Solid на unyly.org, выбери вкладку своего клиента (Claude Desktop, Claude Code, Cursor) и нажми Install — конфиг сгенерируется автоматически, без правки JSON.

Похожие MCP

Fetch

Web content fetching and conversion for efficient LLM usage.

автор: Community

Roblox Studio

Enables AI coding tools to control Roblox Studio for workspace exploration, instance manipulation, and script management. It provides tools for playtesting, sce

paralovавтор: paralov

AWS KB Retrieval

Retrieval from AWS Knowledge Base using Bedrock Agent Runtime.

modelcontextprotocolавтор: modelcontextprotocol

Spring AI MCP Server

Provides auto-configuration for setting up an MCP server in Spring Boot applications.

автор: Community

llm-analysis-assistant

A very streamlined mcp client that supports calling and monitoring stdio/sse/streamableHttp, and can also view request responses through the /logs page. It also

xuzexin-hzавтор: xuzexin-hz

MCP-Agent

A simple, composable framework to build agents using Model Context Protocol by [LastMile AI](https://www.lastmileai.dev)

lastmile-aiавтор: lastmile-ai

Spring AI MCP Client

Provides auto-configuration for MCP client functionality in Spring Boot applications.

автор: Community

mcp.natoma.ai

A Hosted MCP Platform to discover, install, manage and deploy MCP servers by [Natoma Labs](https://www.natoma.ai)

автор: Community

MCPHub

Website to list high quality MCP servers and reviews by real users. Also provide online chatbot for popular LLM models with MCP server support.

автор: Community

MCP Servers Rating and User Reviews

Website to rate MCP servers, write authentic user reviews, and [search engine for agent & mcp](http://www.deepnlp.org/search/agent)

автор: Community

Compare Solid with

Не уверен что выбрать?

Найди свой стек за 60 секунд

Автор?

Embed-бейдж для README

Похожее

Все в категории ai