Best Cursor AI Extensions and Rules in 2026: Power Up Your AI Coding Workflow

How we chose these add-ons

Cursor does not have one app store for everything, so a list of picks mixes file formats, protocols, plugins and editor extensions. We scored each candidate on five points, all taken from official documentation:

• →Is it documented by Cursor or by the add-on's own maintainer, so the claim can be checked?
• →Does it change what the AI agent does, not only how the editor looks?
• →What does it cost, and is there a plan restriction?
• →What can go wrong: does it run code, reach the network or read your data?
• →Is the status clear, or does the documentation call it legacy, deprecated or limited?

The add-ons at a glance

1. Project Rules: the first thing to set up

Project Rules are the current way to give Cursor standing instructions. Cursor's rules documentation says they are stored in .cursor/rules, are version controlled and are scoped to your codebase. The file extension matters: the docs say project rules must use .mdc, and that a plain .md file in that folder is ignored by the rules system. Four behaviors are set through frontmatter. A rule can always apply, attach when matching files are in context, be applied by the agent when it judges the description relevant, or apply only when you mention it. Cursor recommends keeping rules under 500 lines and splitting larger ones. The limits: rules do not affect Cursor Tab or other AI features, and a rule that only points to another file does not inline it, since the agent reads that file with its tools when needed. User Rules are global to your environment, and Team Rules are managed from the dashboard on Team and Enterprise plans. Where they conflict, Team Rules take precedence over Project Rules, then User Rules. Who it is for: everyone. The current rules page documents Project Rules and AGENTS.md and does not mention the older .cursorrules file, so we could not confirm its status from Cursor's current docs; new projects should use .cursor/rules.

2. AGENTS.md: instructions that travel between tools

AGENTS.md is a plain markdown file that Cursor's rules page describes as a simpler alternative to .cursor/rules. Cursor supports it in the project root and in subdirectories, and nested files apply automatically to work in that folder, with the more specific file taking precedence. The AGENTS.md site describes the format as an open format for guiding coding agents, lists Cursor among the supporting tools alongside Codex, Jules, Aider, Zed, Windsurf and GitHub Copilot, says it is stewarded by the Agentic AI Foundation under the Linux Foundation, and reports use by more than 60,000 open-source projects. The limit: Cursor describes it as plain markdown without metadata, and file-pattern scoping is documented only for Project Rules. Cursor's rules page states no length limit for it. Who it is for: teams that use more than one coding agent and want a single set of instructions, and monorepos that need per-package guidance.

3. MCP servers: the biggest gain and the biggest risk

MCP servers connect Cursor's agent to outside tools and data. Cursor supports three transports (stdio, SSE and Streamable HTTP). Servers are configured in .cursor/mcp.json for one project or ~/.cursor/mcp.json for every project, and Cursor resolves variables such as ${env:NAME} in the config so secrets can stay out of the file. OAuth is supported for servers that need it, and marketplace entries install from the Customize panel with Add to Cursor. By default Cursor asks for approval before using MCP tools, and you can inspect a tool's arguments first. A good example for builders is Supabase's MCP server: Supabase's docs list a hosted server, feature groups, and options for read-only mode and project scoping, and name prompt injection as the primary attack vector unique to LLMs. The limit: Cursor's docs state no cap on the number of tools, so we cannot tell you where too many servers start to hurt, and they warn that MCP servers can access external services and execute code on your behalf. Who it is for: people who need the agent to read a database, a tracker or documentation, and who will keep approvals switched on.

4. Cursor Marketplace plugins: bundles from reviewed listings

Cursor announced plugins on February 17, 2026. A plugin can bundle skills, subagents, MCP servers, hooks and rules, and the launch set came from partners that include Amplitude, AWS, Figma, Linear and Stripe. Cursor's plugin docs say you find official plugins at cursor.com/marketplace or by searching in Customize, then choose Install with a project or user scope. Community plugins and MCP servers are listed separately on cursor.directory. Cursor's marketplace security page says every marketplace plugin is manually reviewed before listing, that plugins must be open source, that each update is reviewed, and that plugins are not updated automatically from source. It also says installation is at the discretion and risk of the installing user and recommends reading a plugin's source first. The limit: the page does not describe the review method or claim it catches every problem. Cursor's blog lists private team marketplaces as coming soon, while the plugin docs now mention team marketplaces on Teams and Enterprise. Who it is for: anyone who wants an MCP server and matching instructions installed together.

5. Editor extensions from Open VSX: ESLint, Prettier and a few more

Cursor's extensions documentation says it gets third-party extensions from Open VSX. Most popular extensions are there, but not every Microsoft Marketplace extension is listed, and Cursor publishes audited first-party Anysphere replacements for some that are unavailable. Cursor routes search and downloads through its own proxy and says it runs automated malware and supply-chain analysis before an extension appears. Four extensions we checked on Open VSX, all marked verified by the registry on November 7, 2026: ESLint by dbaeumer (MIT licence, version 3.0.34, about 6.6 million downloads), Prettier by esbenp (MIT, version 12.4.0, about 9.8 million downloads), Error Lens by usernamehw (MIT, version 3.30.0, about 1.4 million downloads) and Tailwind CSS IntelliSense by bradlc (MIT, version 0.16.0, about 2.7 million downloads). They help in a simple way: the agent writes code, the linter and formatter flag and fix what does not match your standards. The limit: Cursor's docs say the same publisher.extension name can point to different code on Open VSX, so treat extension IDs like dependencies. Who it is for: any project with a lint or format configuration.

6. Agent Skills and commands: reusable procedures

Cursor's skills documentation describes Agent Skills as an open standard. A skill is a folder with a SKILL.md file whose frontmatter needs a name matching the folder and a description. Cursor loads project skills from .agents/skills and .cursor/skills, user skills from the matching folders in your home directory, and also reads skills from .claude and .codex folders. The agent decides when a skill is relevant, or you run one by typing a slash in Agent chat. The limits: only ~/.cursor/skills syncs to cloud agents, and only when you turn on Sync Skills for Cloud Agents in Settings; automation runs do not load synced skills. The same page covers a built-in /migrate-to-skills skill (Cursor 2.4 and later) that converts slash commands, and rules set to Apply Intelligently, into skills. We could not find a separate commands page to verify details. Who it is for: teams repeating the same procedure, such as writing tests or preparing a release.

7. Hooks: scripts that can stop the agent

Hooks are spawned processes that talk to Cursor over stdio with JSON, and they can observe, block or change agent behavior. They are defined in hooks.json at project level (checked in with the project), user level, team level (cloud distribution, Enterprise only) or enterprise level, with enterprise settings taking priority. Events include sessionStart, preToolUse, beforeShellExecution, beforeMCPExecution, afterFileEdit and beforeSubmitPrompt, among others. Cursor's documented uses include running formatters after edits, scanning for secrets or personal data, and gating risky operations such as SQL writes. Exit code 2 blocks an action, while other failure codes let the action proceed unless failClosed is set. The limits: cloud agents run command-based hooks only, user-level hooks do not apply to them, and cloud distribution is Enterprise only. Hooks execute code on your machine, so read any hook before you add it. Who it is for: teams that need a hard rule, such as never running a write query, that a polite instruction cannot guarantee.

8. The Claude Code extension: a second agent in the same editor

Anthropic's Claude Code for VS Code is listed on Open VSX, and on November 7, 2026 the registry showed it as verified, at version 2.1.319, with about 64.9 million downloads and an all-rights-reserved licence. In a December 14, 2025 thread on Cursor's community forum, one user says they use it when Cursor's agent terminal misbehaves. The limit: it is a separate product with its own terms and costs, which we did not verify here, and running two agents on one repository means two sets of edits to review. Who it is for: developers who already use Claude Code in the terminal and want it beside the editor. Our Claude Code vs Cursor page compares the two.

Which to set up first

Start with one Project Rule file and an AGENTS.md only if you use other agents too. Rules cost nothing and reduce repeated explanations. Second, install the linter and formatter extensions your project already uses, so AI-written code is checked by the same tools as human code. Third, add one MCP server for a real need and keep manual approval on. Fourth, turn repeated procedures into skills. Leave hooks until you have a rule that must never be broken. If you are still choosing an editor, see our best AI code editor page, and for a full workflow read our vibe coding with Cursor guide.

What it costs

Rules, AGENTS.md, skills, hooks and MCP are features of Cursor itself, and the pages we read attach no separate price to them, except that Team Rules need a Team or Enterprise plan and hook cloud distribution is Enterprise only. As of November 2026, Cursor's pricing page lists Free at $0 with limited agent and chat requests, Plus at $8 a month limited to Grok models, Premium at $30 a month with four times Plus usage and third-party models billed separately at API list price, Super at $100, Ultra at $200 and Teams from $30 per user a month. The four extensions named above (ESLint, Prettier, Error Lens, Tailwind CSS IntelliSense) are MIT licensed on Open VSX; the Claude Code extension is not, and Cursor's pages give no prices for individual MCP servers or plugins. See Cursor for our listing, or Cursor alternatives if the price does not fit.

Security risks of MCP servers, plugins and extensions

Every add-on in this list either runs code or steers an agent that can. These points come from Cursor's and Supabase's own documentation:

• →Cursor's docs say to install MCP servers only from trusted developers and repositories, because they can access external services and execute code on your behalf.
• →Prompt injection is real. Supabase's docs say text in your data, such as a support ticket, can trick an assistant into running harmful queries, and that its protective wrapping is not foolproof.
• →Keep per-tool approval on. Cursor asks before using MCP tools by default, and its security page says run modes are best-effort guardrails, not a hard security boundary.
• →Scope servers narrowly. Supabase documents read-only mode, project scoping and feature groups, and advises connecting to production only when the task needs it.
• →Marketplace plugins are reviewed by hand, but Cursor still says installation is at your own risk and recommends reading the source.
• →Extension names are not identity. Cursor says the same publisher.extension name can refer to different code on Open VSX, and offers install cooldowns and publisher allowlists to administrators, plus signature checks for Enterprise teams.
• →Workspace trust exists but is off by default in Cursor, and Cursor says restricted mode breaks AI features. Do not open untrusted repositories with an agent that can run commands.

What changed recently and what we do not know

Recent facts, then gaps:

• →February 17, 2026: Cursor introduced plugins and the Cursor Marketplace.
• →The extensions page describes Open VSX screening through a Cursor proxy, admin install cooldowns and, for Enterprise teams, signature checks.
• →Cursor's changelog entries from October 13 to November 6, 2026 cover an app marketplace, the Agent SDK, cloud agents, live apps and voice chat, and none mentions rules, AGENTS.md or .cursorrules, so we found no recent rule changes.
• →Unclear: the current rules page does not mention .cursorrules, and we could not confirm any official deprecation notice or end date.
• →Unknown: how many MCP tools Cursor handles well (its MCP page states no cap), and how the plugin review works in detail. The hooks page does not label hooks as beta.
• →Unknown: Anthropic's plan requirements for the Claude Code extension, and exact prices of third-party plugins.
• →We have not seen independent data on how much any add-on improves code quality.

How we chose, and what we did not do

We compared Cursor's rules, MCP, plugins, skills, hooks, extensions, security and pricing documentation, the AGENTS.md and Supabase docs, and Open VSX listings on November 7, 2026. We did not test these add-ons ourselves, so we make no claims about speed or code quality. Vibe Code Apps is an independent directory with no affiliation to Cursor. Prices and features change, so check the linked pages before you pay. For wider context, see Cursor AI, the risks of vibe coding and agentic AI coding tools, or compare Cursor vs Windsurf.

Our verdict

Write a Project Rule first, add AGENTS.md if you mix agents, and connect one MCP server only with read-only scope and manual approval. Install ESLint and Prettier so a human standard checks AI output. Add plugins, skills and hooks as needs appear, not before. Nothing here replaces reviewing the code Cursor writes.

Apps and tools mentioned

1

CursorVerified

Cursor is an AI code editor, CLI and cloud agent product that supports Project Rules, AGENTS.md, MCP servers, plugins, skills and hooks. As of November 2026 Free is $0, Plus is $8 a month (Grok models only), and Premium is $30 a month with third-party models billed separately at API list price.

cursor

Visit site →

2

Claude CodeVerified

Anthropic's coding agent, also available as an extension listed on Open VSX that can run beside Cursor's own agent. Check Anthropic's terms for plan requirements.

other

Visit site →

5.5

vibe

3

HireTrack AI

An applicant tracking system listed in our directory as built with Cursor and Supabase.

cursor

supabase

Visit site →

4.6

vibe

4

Schedsy

A scheduling and booking tool listed in our directory as built with Cursor and Supabase.

cursor

Visit site →

2.3

vibe

5

AgentBoard

An agent monitoring tool listed in our directory as built with Cursor and Claude Code.

cursor

claude-code

Visit site →

2.4

vibe