Best Cursor Plugins in 2026

The best Cursor plugins in 2026 are UI Craft, Mission Kit, NotFair, spec-superflow, Blender Developer Tools, CometChat, Firecrawl, LMCP, PlayCanvas and Assay. Each one ships a Cursor plugin manifest, has no high or critical finding in the HOL Plugins catalog, and documents a Cursor install path in its own repository. For each pick we opened .cursor-plugin/plugin.json and checked it against Cursor's plugin reference, so you know which rules, skills, agents, commands, hooks and MCP servers will load before you install.

The Best Cursor plugins page on hol.org ranks every Cursor listing by Trust Score. This list is ordered differently. Plugins that use more of what Cursor supports, such as .mdc rules and Cursor hooks, come first, and plugins that mostly hand Cursor a folder of skills come later. A plugin can rank high here and lower on the hub, or the reverse.

TL;DR: best Cursor plugins

Labels and scores were read from the HOL catalog on Nov 2, 2026, and star counts come from HOL's GitHub fetch the same day. HOL had no star count on record for Firecrawl's skills repository.

How we picked these

HOL lists 208 plugins that name Cursor as a supported runtime. Only 14 of them name Cursor as their primary runtime. The rest list another runtime first, most often Codex, Claude Code or MCP. HOL's scan found a .cursor-plugin manifest in 151 of the 208, and 34 of those also ship a .cursor-plugin/marketplace.json, the file Cursor needs to import a repository from GitHub. By security label, 68 are Safe, 111 Review and 23 Caution, and 6 had not been scanned when we read the catalog.

To make this list, a plugin had to ship a Cursor manifest, carry no high or critical finding, and show only repository hygiene findings above the info level: a missing lockfile, Dependabot config or SECURITY.md. It also had to document a way to get it into Cursor somewhere in its own repo. We took one plugin per job and left out plugins already covered in our Codex and Claude Code roundups.

Then we read each manifest. Cursor's plugin reference says a field in plugin.json replaces folder discovery for that component, and that without the field Cursor looks in default places: skills/, rules/, agents/, commands/, hooks/hooks.json and mcp.json. That check turned up two things a star count won't show you: Assay's Cursor hooks file is empty by design, and PlayCanvas suggests a symlink install that Cursor skips. Both are covered below.

One caveat applies to every entry. On all ten, HOL's per-skill content scanner did not complete, so the scans are partial. HOL checked manifests, MCP settings, provenance and the supply chain, but it did not read each SKILL.md in depth, and the scan is static. Read the label as evidence about packaging, not as a code review. The Trust Score page explains how the labels are set.

What are Cursor plugins?

A Cursor plugin is a Git repository, or a folder in one, that Cursor installs as a single unit. Cursor loads two formats. A Cursor Plugin has a .cursor-plugin/plugin.json manifest and can carry six kinds of component:

• Rules: .mdc files with standing instructions, applied always or by file glob.
• Skills: folders with a SKILL.md that the agent loads when a task matches.
• Agents: markdown files that define custom agents.
• Commands: files the agent can run as slash commands.
• Hooks: scripts that run on events such as sessionStart, beforeShellExecution or afterFileEdit.
• MCP servers: tool connections, configured in mcp.json at the plugin root.

The second format, an Agent Plugin, follows the open Agent Plugins standard. It has a plugin.json at the root and can only package skills and MCP servers. Several picks below ship both. According to Cursor's plugin docs, plugins work in the desktop app, on the web and in the CLI. Plugins in the official Cursor Marketplace are open source and reviewed by hand before they are listed. That review covers marketplace listings, not repos you import or clone yourself.

How to install Cursor plugins

In the Cursor app

Open Customize in the sidebar, find the plugin and select Install, then choose project or user scope. To bring in a plugin that isn't in the marketplace, import its GitHub repository from Customize. That route needs a .cursor-plugin/marketplace.json in the repo. Teams and Enterprise admins can also add a repo as a team marketplace from Dashboard, then Plugins & MCPs, and set each plugin to Default Off, Default On or Required.

From a local folder

Copy or clone the plugin into ~/.cursor/plugins/local/, then restart Cursor or run Developer: Reload Window, and open Customize to check that its components loaded. Cursor skips a symlink in that folder if it points to a directory outside it, so clone or copy instead of linking. On Enterprise plans, local plugins load only when an admin turns on Allow Local Plugin Imports, which is off by default.

In the Cursor CLI

Inside an agent session, /plugin lists, installs and uninstalls plugins at user or project scope, and /plugin marketplace add adds a marketplace by git URL. From your shell, you can manage marketplaces without opening a session:

agent plugin marketplace add https://github.com/owner/repo
agent plugin marketplace list

Add --git-ref to pin a branch, tag or commit. To try a plugin for one session without installing it, point the CLI at a checkout:

agent --plugin-dir ./my-plugin

Cursor skills vs plugins

Cursor also reads skills from .claude/skills/ and .codex/skills/, so a skill you already use with Claude Code or Codex works in Cursor without a plugin. Rules and hooks are where a Cursor plugin does more than a skill can: a skill waits for the agent to decide it is relevant, while a rule with alwaysApply or a hook on beforeShellExecution applies whatever the agent decides.

The 10 best Cursor plugins

1. UI Craft: design rules and a check before every file write

Eduardo Calvo's design system uses all six component types Cursor supports, more than any other pick here. Its Cursor manifest loads four skills (the main one plus minimal, editorial and dense-dashboard variants), 25 command files such as /craft, /critique and /polish, two agents named design-reviewer and a11y-auditor, four .mdc rules, and an MCP server pinned to [email protected]. The hook is a preToolUse entry that matches writes and runs UI Craft's detector before the agent saves a file. The README frames the project as a ladder: install and keep prompting as usual, then add commands, saved design context and finally gates in review and CI.

• Trust Score 78, security 88 (Review), 0 high findings
• Finding behind the label: no Dependabot config in the repo
• 421 GitHub stars; category Agent Skills

Install from the README, then reload the window:

git clone https://github.com/educlopez/ui-craft.git ~/.cursor/plugins/local/ui-craft

Example prompts from the README. /start reads your framework, tokens and existing UI and tells you which step to take next:

/craft dashboard
/start

Pros: the hook gives you a check the agent can't skip, and the critique output scores pages against Nielsen's ten heuristics and six design laws, with each finding tagged by business impact.

Cons: the hook runs a Node script on every write, so you need Node installed. The listing is not owner verified on HOL. Teams and Enterprise accounts need local plugin imports allowed before the clone loads.

HOL listing · GitHub

2. Mission Kit: plans you approve, production pushes you confirm

Mission Kit calls Cursor its first-class surface, and its manifest shows why. It points Cursor at 25 rules, 10 core skills, 14 agents and 29 commands under .cursor/, plus a hooks file with five Cursor events. Among them, beforeShellExecution runs a script called guard-shell.sh and beforeSubmitPrompt runs secrets-prompt.sh. You describe a goal, approve the plan and the first unit, and the agent works one unit at a time. Promotion to main only happens through /git-prod, after you say yes, and direct commits to main are blocked. Confirmations use Cursor's Ask questions.

• Trust Score 95, security 93 (Safe), 0 high findings
• Finding behind the label: no Dependabot config
• 22 GitHub stars; owner verified on HOL, with Scanner Clean and Marketplace Ready badges

Install from your project root, per the README (Node.js 22.12 or newer):

npx @dadado/agent-kit-cli@latest install

Example prompts from the README:

/start-project
/continue-plan

Pros: the highest Trust Score on this list, and the only pick that hooks Cursor's shell and prompt events. /continue-plan picks up in a fresh chat from saved resume state.

Cons: 22 stars is a small user base. The install writes its commands, rules and agents into your project's .cursor/ folder, and the staging and production commands assume a git flow with a staging branch. The README says VS Code and Windsurf get only partial support.

HOL listing · GitHub

3. NotFair: SEO, GEO and ad accounts from one MCP connection

The most-starred pick ships 48 skills for site audits, keyword research, GEO work, GA4, Search Console and paid media on Google, Meta, X, LinkedIn, Reddit and TikTok. Its Cursor manifest declares the skills folder and a .mcp.json with one hosted NotFair MCP server, which you sign in to through a browser OAuth flow when Cursor prompts you. Each skill has one SKILL.md entry point that loads a readable WORKFLOW.md, so you can see what the agent will do before it does it.

• Trust Score 77, security 88 (Review), 0 high findings
• Findings behind the label: no lockfile for requirements.txt, no Dependabot config
• 4,095 GitHub stars, 513 forks; category Tools & Integrations

The README says to install NotFair from Cursor's plugin marketplace. For local testing it gives this, followed by a Cursor restart:

git clone https://github.com/nowork-studio/notfair-plugin.git ~/.cursor/plugins/local/notfair

Example prompts from the README:

Audit my site and tell me why organic traffic fell.
Find pages that could earn citations in AI answers.

Pros: one connection covers ad platforms, analytics, WordPress and GoHighLevel, and the workflows are plain files you can read and edit.

Cons: the skills can make changes in connected ad accounts and WordPress sites, and that access runs through NotFair's hosted server. Not owner verified on HOL.

HOL listing · GitHub

4. spec-superflow: a change counts as done only when the check passed

MageByte's workflow gives each change one of two paths. Direct is for a bounded fix with a known check. Planned is for work that needs agreement first, and it writes a short proposal.md and tasks.md that you approve once. Completion records the verification command, its exit code and the git range that was reviewed. A failed check stays in execution for repair, and shipping with a known issue is recorded as accepted-risk. The project's INSTALL guide says the Cursor installer copies the 9 skills into .cursor/skills/ and writes .cursor/rules/phase-guard.mdc with alwaysApply: true.

• Trust Score 93, security 92 (Safe), 0 high findings
• Finding behind the label: no Dependabot config
• 879 GitHub stars, 92 forks; owner verified on HOL

Install for Cursor, from the README (Node.js 20 or newer):

npx spec-superflow@latest install-cursor

The README has no chat prompts; it drives work through the ssf CLI. Two examples from it, a direct fix and resuming a change:

ssf workflow start changes/fix-login-timeout --path direct --scope "Fix login timeout without changing the authentication protocol"
ssf resume changes/example

Pros: the CLI uses only Node's standard library at runtime, and the README says defaults never record a pass that didn't happen. Subagents and worktrees are opt-in.

Cons: the main README is in Chinese; the English one is at docs/README_en.md. The phase-guard rule applies in every session, which adds to the context of each chat.

HOL listing · GitHub

5. Blender Developer Tools: rules that catch dead bpy APIs

This is the clearest case on the list for Cursor rules. The plugin ships 25 skills and 12 .mdc rules, and Cursor applies each rule wherever its scope glob matches, so a rule such as no-bgl.mdc or always-free-bmesh.mdc applies even when the agent hasn't loaded a skill. The README lists what the rules target: operator overrides through bpy.context.copy(), which Blender removed in 4.0; action.fcurves on Blender 5's slotted actions; and glTF exports that land on their back in every engine. Each contract is backed by a headless example that CI smoke-tests on Blender 5.2 LTS and 4.5 LTS.

• Trust Score 82, security 88 (Review), 0 high findings
• Finding behind the label: no Dependabot config
• 19 GitHub stars; owner verified on HOL

Install from the slim plugin-dist branch, per the README, then run Developer: Reload Window:

git clone --branch plugin-dist --single-branch https://github.com/TMHSDigital/Blender-Developer-Tools.git ~/.cursor/plugins/local/blender-developer-tools

The README has no sample prompts. It says the agent picks up a skill when its description matches the task, and Customize should list the 25 skills and 12 rules after the reload.

Pros: the rules work without the agent choosing them, and a second, opt-in plugin, blender-live-session, connects the official Blender Lab MCP server for a running Blender 5.1 or newer.

Cons: the written content is licensed CC BY-NC-ND 4.0, with only code snippets and templates under MIT. The README warns that the live-session server runs any code the agent sends, without guards.

HOL listing · GitHub

6. CometChat: chat and calling added to the app you already have

CometChat's official pack covers six UI kits (React v7, Angular v5, React Native v5, iOS v5, Android v6 and Flutter v6) plus headless Calls SDK skills. Its Cursor manifest declares the skills folder and an inline MCP server at https://mcp.cometchat.com/mcp, so the skills read current CometChat docs as they work. A cometchat dispatcher skill asks about your framework, placement and credentials, gets your approval on a plan, then hands off to the right platform skill.

• Trust Score 82, security 89 (Review), 0 high findings
• Findings behind the label: no lockfile for package.json, no Dependabot config
• 161 GitHub stars; owner verified on HOL

Install the skills for Cursor with the README's installer, which detects your framework:

npx @cometchat/skills add --ide cursor

Example prompts from the README:

add chat to my app
migrate my app to CometChat

Pros: the migration prompt detects Sendbird or Stream Chat and does the work on a cometchat-migration git branch. It comes from the vendor, so the skills track CometChat's own task guides.

Cons: the npx route installs skills only; the README says the docs MCP comes bundled with a plugin install. Every skill assumes CometChat as your provider.

HOL listing · GitHub

7. Firecrawl: web scraping for the session or for your codebase

Firecrawl's skills repository has a Cursor manifest that bundles two of its three skill sets: 13 core skills for live search, scrape, crawl, map and interact during a session, and 5 build skills for adding Firecrawl's API to product code. The README says the workflow recipes are left out of the plugin because they are session recipes. The manifest doesn't declare MCP servers, so Cursor finds the root mcp.json, which starts firecrawl-mcp with a FIRECRAWL_API_KEY placeholder. The repo also has a .cursor-plugin/marketplace.json, so the Customize GitHub import and agent plugin marketplace add both work on it.

• Trust Score 88, security 88 (Review), 0 high findings
• Finding behind the label: no SECURITY.md
• GitHub stars: not recorded in HOL's catalog; owner verified on HOL

Install from the README, which opens a picker over the whole catalog:

npx skills add firecrawl/skills

Example requests, from the README's guide to which skills you need:

Search/scrape the web for me right now, during this session
Add Firecrawl to this codebase

Pros: one repo covers the agent using the web now and your app using it later, and the skills are CI-synced copies from the CLI and SDK repos where Firecrawl maintains them.

Cons: you need a Firecrawl account and API key, or a self-hosted deployment. Cursor's submission checklist asks plugins to declare every ${VAR} in mcp.json in the manifest, and Firecrawl's Cursor manifest doesn't declare the key. HOL's maintenance score for this listing is 40.

HOL listing · GitHub

8. LMCP: your Mail, Calendar and Office files, through a local server

LMCP runs an MCP server on your own Mac or Windows machine and connects the agent to apps such as Apple Mail, Outlook, Teams, Slack, Notion, OneDrive, Word and Excel. The repo doubles as a Cursor plugin. It adds the local-mcp server through mcp.json, a rule that tells the agent when to use LMCP, and 14 skills for email, calendar, messages, files and Office. The plugin only starts the server; the LMCP desktop app has to be installed first.

• Trust Score 80, security 94 (Safe), 0 high findings
• Finding behind the label: no Dependabot config
• 75 GitHub stars; Scanner Clean and Marketplace Ready badges

Install the LMCP app from its download page, then, per the README, either install the Cursor plugin from the repository or run:

npx -y local-mcp@latest setup

Example prompts from the README:

Help organize my day so I can meet my commitments.
Prepare replies to customer orders using the inventory and plan dispatch.

Pros: the tools run on your machine with no API keys to manage, and the README asks you to review proposed actions before you authorize them.

Cons: the core binary is proprietary; this repo holds docs, the npm wrapper and the plugin. Apple-native tools need macOS. Not owner verified on HOL.

HOL listing · GitHub

9. PlayCanvas: 3D apps checked in the browser, not just built

PlayCanvas's official skills cover its Engine, React and Web Components surfaces, and the skills work out which one your project uses from its imports before reading the matching reference. There are 16 skills, including inspect-glb, which measures a model's bounds, joints and clips offline, and verify-pixels, which proves a rendering change is identical or within bounds. The README badge puts the startup cost at about 700 tokens of context.

• Trust Score 78, security 83 (Review), 0 high findings
• Findings behind the label: no lockfile for package.json, no SECURITY.md, no Dependabot config
• 32 GitHub stars; owner verified on HOL

For a single CLI session, the README loads the plugin from a checkout. It uses the older cursor-agent name; Cursor's current CLI docs call the command agent:

cursor-agent --plugin-dir .

Example prompts from the README:

Build a Direct Engine product viewer for these GLBs. Inspect and calibrate every model, reuse the shipped camera controls, and verify grounding and framing in the browser.
Polish this @playcanvas/react prototype with compressed assets, reused Engine scripts, pooled effects, deliberate lighting and tone mapping, and screenshot checks.

Pros: the skills check the package versions installed in your project before using an API, and finish with runtime and screenshot evidence.

Cons: for a persistent install the README suggests linking the repo into ~/.cursor/plugins/local/playcanvas, but Cursor skips symlinks that point outside that folder, so clone or copy it there instead. The README also warns that installing both the skills and the plugin for one agent shows every skill twice.

HOL listing · GitHub

10. Assay: presses every button on the page the agent just changed

Assay opens a page the agent wrote, drives every control on it and reports in one line, clean or not, for example assay: checked todo/todo.html, 8 checks, nothing flagged. In Claude Code, Codex and four other harnesses, the plugin also adds a Stop hook that runs the check at the end of every turn that changed a page. Cursor is the exception: the repo's install guide says Cursor's hooks take a different format, and the hooks-cursor.json the manifest points to is empty. In Cursor you get the skill, and the agent decides when to run it.

• Trust Score 92, security 95 (Safe), 0 findings above info
• 109 GitHub stars; owner verified on HOL, Scanner Clean badge

Assay needs its CLI on your PATH first (pip install assay-ui, Python 3.10 or newer). Then, from the install guide, in the Cursor CLI:

agent plugin marketplace add https://github.com/awss1i/assay

Then install assay from /plugin in a session. The same guide gives /add-plugin assay@https://github.com/awss1i/assay for Agent chat in Cursor 2.5 and later. The README has no chat prompts; the skill tells the agent to run this after it finishes a page and paste the output word for word:

assay <the page> --one-line

Pros: the cleanest scan of the ten. It reports and doesn't fix, and the README says a finding is a place to look rather than a confirmed bug.

Cons: without the hook, Cursor users rely on the agent remembering to run it. The assay CLI needs Python 3.10 or newer, or uv to fetch one.

HOL listing · GitHub

Popular Cursor plugins we left out

claude-octopus has the most stars of any listing that names Cursor, at 4,357, but HOL flags four high findings for hardcoded secrets and gives it a Trust Score of 35. LetsFG, with 2,141 stars, is one of the 14 listings that name Cursor as their primary runtime, yet the scanned commit has no Cursor or Agent Plugins manifest; it connects through a remote MCP server instead. Aegis (1,411 stars) ships a .cursor-plugin manifest, but its README gives no Cursor install guide, and HOL raises two medium findings about how its package.json is set up for DeepSeek Harness. Digital Marketing Pro (922) covers the same ground as NotFair, Uizze overlaps with UI Craft, and better-harness (2,430) was in our Codex list. If you want the catalog ranked by Trust Score instead, the Best Cursor plugins page does that, and humanizer-ru, a Russian text editor scoring 98, sits at the top.

Build your own Cursor plugin

Start from Cursor's plugin template repository. A Cursor Plugin manifest needs only a name, in lowercase kebab-case. Put components in the default folders, or point to them from plugin.json, but remember that a manifest field replaces the default folder for that component. To test, place the plugin in ~/.cursor/plugins/local/ and reload the window, or run agent --plugin-dir ./my-plugin. If the repo holds more than one plugin, or you want people to import it from GitHub, add a .cursor-plugin/marketplace.json at the root. If your mcp.json uses ${VAR} placeholders, declare each one under variables in the manifest. Submit at cursor.com/marketplace/publish for Cursor's review.

Before you publish, run the HOL Plugin Scanner on the repo. It can also run in GitHub Actions. Most labels in this list would go up with a committed lockfile, a SECURITY.md and a Dependabot config. Then submit the plugin to HOL Plugins for a public Trust Score and listing. To review what installed plugins do on your machine at runtime, HOL Guard checks agent actions before they run.

What do people ask about Cursor plugins?

What is the best Cursor plugin in 2026?

It depends on the job. UI Craft uses the most of what a Cursor plugin can carry: skills, commands, agents, rules, a hook and an MCP server. NotFair has the most GitHub stars of the ten picks at 4,095, and Assay has the highest security score at 95. For a list ranked by Trust Score alone, see the Best Cursor plugins page on hol.org.

How do I install a plugin in Cursor?

Open Customize in the sidebar, find the plugin, select Install and choose project or user scope. For a plugin outside the marketplace, import its GitHub repository from Customize if it has a .cursor-plugin/marketplace.json, or clone it into ~/.cursor/plugins/local and reload the window. In the CLI, use /plugin inside a session or agent plugin marketplace add with the repo URL.

Do Cursor plugins work in the Cursor CLI?

Yes. Cursor's docs say plugins work in the desktop app, on the web and in the CLI. In the CLI, /plugin installs and manages them, and agent --plugin-dir loads a local plugin folder for one session.

What is the difference between a Cursor rule, a skill and a plugin?

A rule is an .mdc file with instructions Cursor applies always or to files that match a glob. A skill is a folder with a SKILL.md that the agent loads when a task matches or when you call it with /. A plugin is a package that can bundle rules, skills, agents, commands, hooks and MCP servers and install them together.

Can I use Claude Code skills in Cursor?

Yes. Cursor loads skills from .claude/skills/ and .codex/skills/ as well as its own .cursor/skills/ and .agents/skills/ folders. Claude Code plugins are a different package format, and several picks here ship a Claude manifest and a Cursor manifest side by side.

Are Cursor plugins safe to install?

Not automatically. A plugin can include hooks that run scripts on events such as shell execution, and MCP servers that run as you. Cursor reviews marketplace plugins by hand before listing them; that review does not cover repos you import or clone yourself. Read the manifest, check the plugin's HOL listing for its label and the findings behind it, and remember that HOL's scan is static and partial for these ten.

Why does a plugin's hook not run in Cursor?

Cursor hooks use their own event names and file format, so a hook written for another agent may not carry over. Assay is an example: its plugin adds a Stop hook in Claude Code and Codex, but its Cursor hooks file is empty, so Cursor users get the skill only.