Best Grok Plugins in 2026

The best Grok plugins in 2026 are Antidote, go-ultimate, Dryforge, Digital Marketing Pro, Keenable, Cargo, Homie, Second Claude Code, telepathy and Alcove. All ten are listed in the HOL Plugins catalog with no high or critical scan finding, and each README explains how to install it in Grok Build, xAI's coding agent. They are sorted by how much each one can do on your machine once you trust it, starting with plugins that only add instructions and ending with one that installs a program when a session starts.

The order matters because Grok Build asks for trust once, at install. The --trust flag switches on a plugin's skills, hooks and MCP servers all together, and none of these ten is in xAI's official marketplace, so the decision is yours alone. The Best Grok plugins page on hol.org ranks all 68 Grok listings by Trust Score. This post covers the step after that: what each pick switches on when you say yes.

TL;DR: best Grok plugins

Scores, labels and findings were read from the HOL catalog API on Nov 9, 2026. The number in brackets is HOL's security score. Star counts come from HOL's GitHub fetch the same day. Component counts come from each plugin's Grok manifest and files at the commit HOL scanned.

How we picked these

HOL lists 68 plugins that name Grok as a supported runtime. Only 10 name Grok first. The rest name Codex (18), Claude Code (18), the Agent Plugins format (9), MCP (6), Cursor (4), Gemini CLI (2) or OpenCode (1). HOL's file list shows a .grok-plugin folder in 43 of the 68, and 44 ship a Claude Code manifest, which Grok Build also reads. Of the 68, 20 are labeled Safe, 42 Review and 4 Caution, and 2 have no scan yet.

Two gaps shaped this list. First, HOL's scanner has no Grok-specific checks yet. It validated the Codex, Claude Code, Gemini and Kimi manifests in these repos, but not the Grok one, so we opened each .grok-plugin/plugin.json ourselves at the commit HOL scanned and counted what it loads. Second, xAI's own plugin marketplace lists 36 plugins, and only one HOL Grok listing, HTML/CSS to Image, is among them. Most of what you find in the catalog installs straight from GitHub.

To make this list, a plugin had to have no high or critical finding and give a Grok Build install in its own README. We took one plugin per job and skipped plugins already picked in our Codex, Claude Code, Cursor and Kimi roundups, which rules out HTML/CSS to Image.

One caveat applies to all ten. HOL's per-skill content scanner failed on every pick, 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. Treat the label as evidence about packaging, not a code review. The Trust Score page explains how labels are set.

What are Grok plugins?

"Grok plugin" means three different things depending on which Grok you use.

• Grok Build is xAI's coding agent for the terminal. A Grok Build plugin is a folder that can hold skills, slash commands, agents, hooks, MCP servers and LSP servers. This is what the HOL catalog lists and what this post covers.
• Grok Bot is xAI's agent that works on its own cloud computer. It calls its plugins apps, and you add them from Connect apps in the sidebar. A connected app works for every Bot on your account.
• Grok chat on grok.com uses connectors. Besides the built-in ones you can add your own MCP server as a custom connector at grok.com/connectors.

According to xAI's Grok Build docs, a plugin's manifest goes at .grok-plugin/plugin.json, and it is optional. Without one, Grok finds components in standard places: skills/, commands/, agents/, hooks/hooks.json, .mcp.json and .lsp.json. Grok also reads Claude Code plugins, marketplaces, skills and hooks with no extra setup, which is why so many listings work in Grok without being built for it.

How to install a Grok plugin

From the terminal

Install a plugin straight from GitHub, or add a marketplace first and install by name:

grok plugin install owner/repo --trust
grok plugin marketplace add owner/repo
grok plugin install plugin-name --trust

The source can also be owner/repo#subfolder for a plugin inside a larger repo, owner/[email protected] for a tag, or owner/repo@ followed by a full commit SHA for an exact version. Without --trust, Grok prints the source, warns that installing turns on the plugin's hooks, MCP servers and skills, and stops. Three READMEs below leave the flag off, so expect that warning and run the command again with --trust.

Plugins stay off until they are enabled. If one doesn't show up, check grok plugin list, then run grok plugin enable plugin-name. grok plugin details plugin-name lists what a plugin adds, and grok inspect shows every skill, agent, hook and MCP server loaded in the current folder, each labeled with the plugin it came from.

Inside Grok Build

Type /plugins to open the plugins window, or /marketplace to open it on the Marketplace tab. Press i on a marketplace entry to install it, or press a on the Plugins tab to add one from owner/repo, a URL or a local folder. Press r to reload, or start a new session, to pick up a new plugin. MCP servers show as blocked while a plugin is untrusted.

In Grok Bot and Grok chat

Grok Bot uses its own app catalog rather than grok plugin install. Open Connect apps, find the app, click Add and sign in. A Bot can also add an MCP server that isn't in the catalog when you give it the address. In Grok chat, open grok.com/connectors, choose New Connector, then Custom, and enter an MCP server URL. Cargo, Homie and telepathy below each document a Grok Bot route.

Grok skills vs plugins

A skill can only tell Grok how to do something. A plugin can also give it a tool to do it with, and a hook that runs before or after the agent acts. That is the line this list is sorted on.

The 10 best Grok plugins

1. Antidote: stop the agent from papering over bugs

Antidote is one skill that pushes back on a familiar agent habit: wrapping a bug in null checks, swallowed errors and workarounds until it stops showing. It tells the agent to validate data once at the edge, use the right type or schema, and remove the cause. It has a fix mode for one bug and an audit mode for a whole codebase. The README says it is pure prompt, with no scripts, no network calls and no dependencies.

• Trust Score 84, security 88 (Review), 0 high findings
• Findings behind the score: three low ones, all a missing Dependabot config. HOL takes 10 percent off because the repo doesn't run its scanner in CI, then adds 2 percent for linking its HOL badge.
• 25 GitHub stars; owner verified on HOL
• What Grok loads: one skill from skills/. HOL counts two because the repo carries a second copy for GitHub Copilot.

Install, from the README:

grok plugin marketplace add Avtr99/antidote
grok plugin install antidote@avtr-antidote

Or directly: grok plugin install Avtr99/antidote. Add --trust to either install command. Example prompts from the README:

use antidote
audit for overcomplication

Pros: the least you can hand a plugin. Trusting it turns on one instruction file and nothing that runs. The README reports benchmark results across five models.

Cons: the README's install commands leave off --trust. Its usage section names /antidote for Claude Code, Codex, Cursor and Devin but not Grok. xAI's docs say plugin skills appear in Grok's slash menu, so look for it there or ask in plain words.

HOL listing · GitHub

2. go-ultimate: one set of Go rules the agent follows every time

go-ultimate turns on whenever a task touches Go. It works out what kind of project you have (script, CLI, library, backend service, MCP server or AI agent), loads the matching guide and applies the same review checklist. It pins the Google and Uber style guides as sources and says plainly that it overrides generic Go advice when they conflict. Six smaller skills come along for minimal changes, pull request cleanup, test audits and technical docs.

• Trust Score 92, security 95 (Safe), only info-level findings
• 31 GitHub stars; owner verified on HOL (manual review); Scanner Clean badge
• What Grok loads: 7 skills, no hooks, no MCP server

Install, from the README:

grok plugin install 'Djarvur/cc-mplace#go-ultimate' --trust

Example prompts from the skill's own test cases:

Scaffold a new HTTP service called 'greeter' that exposes GET /hello?name=X and returns a greeting.
Review this Go function.

Pros: the highest Trust Score in this list. Before it uses a language feature, it checks your project's Go version with a small bundled script, so you don't get syntax your toolchain can't build.

Cons: xAI's docs say the part after # is a folder inside the repo, and Djarvur/cc-mplace has no go-ultimate folder. It is a marketplace that points to the go-ultimate repo. If the README command fails, run grok plugin marketplace add Djarvur/cc-mplace and then grok plugin install go-ultimate --trust. That marketplace pins a newer commit than the one HOL scanned. The rules are strongly opinionated, so a team with its own Go style will disagree with some of them.

HOL listing · GitHub

3. Dryforge: agree on the plan before any code is written

Dryforge splits work into two commands. /ready takes whatever you have, from a one-line idea to a PDF spec, asks only the questions it can't answer from your project, and writes the agreed intent as Markdown under .dryforge/. Nothing is built until you approve it. /go then builds it, checks it with tests it runs, and on an existing project works on its own branch and asks how to merge. A third command, /migration, brings an existing codebase in once.

• Trust Score 71, security 80 (Review), 0 high findings
• Findings behind the score: two low ones, a missing SECURITY.md and a missing README in the folder HOL scanned. HOL takes 10 percent off for no scanner in CI.
• 444 GitHub stars, 39 forks, the most starred pick here after Digital Marketing Pro
• What Grok loads: 3 skills. HOL scanned the repo's agent-plugin/ folder, but Grok installs from a grok/ folder. We compared them: the 32 non-manifest files are identical.

Install, from the README (add --trust to the second line):

grok plugin marketplace add prekuter/dryforge
grok plugin install dryforge

Example prompts from the README:

/ready <anything>
/go

Pros: skills only, so trusting it turns on no hooks and no servers. What it writes is plain Markdown in your repo that any agent can read, and the README says those files stay if you remove the plugin.

Cons: the lowest Trust Score of the ten. The README says Grok auto-update is on, so new releases load at the start of each session without a step from you, and its marketplace entry points at a folder rather than a pinned commit.

HOL listing · GitHub

4. Digital Marketing Pro: an agency's playbook in one install

Digital Marketing Pro is the biggest plugin here by a wide margin: 164 skills and 24 specialist agents for brand setup, strategy, competitor research, ad platforms, search and AI-answer visibility, and compliance work such as EU AI Act labeling. It is built for agencies and in-house teams running many brands, and its README describes a 12-part strategy flow that produces a stack of planning documents per brand.

• Trust Score 81, security 86 (Review), 0 high findings
• Findings behind the score: three medium ones for a package.json with no lockfile, plus three low Dependabot notes. HOL takes 10 percent off for no scanner in CI.
• 939 GitHub stars, 153 forks; owner verified on HOL
• What Grok loads: the Grok manifest points at skills/ only, so agents, commands and hooks/hooks.json load from their standard folders. That hooks file holds a check that gates live writes before a shell command runs.

Install, from the README (it says to append --trust):

grok plugin install indranilbanerjee/digital-marketing-pro

The README notes that outside Claude Code you ask in plain language rather than typing its slash commands. Example prompts from the README:

Let's set up a brand for ACME Corp
Run a competitor analysis on stripe.com

Pros: the widest coverage of any marketing plugin in the catalog, and a write gate so the agent can't push changes to a live account on its own. It ships a native Grok manifest instead of relying on Claude compatibility.

Cons: 164 skills is a lot to trust in one step, and HOL's skill scanner did not read them. The write gate uses if filters from Claude Code's hook format, and we did not test whether Grok honors them. Its scripts need Python.

HOL listing · GitHub

5. Keenable: web search with nothing to sign up for

Keenable is one of the ten plugins in the catalog built for Grok first. It connects Grok Build to Keenable's hosted search server, which finds pages and returns their text as clean Markdown, and adds two skills: one for everyday lookups and one for multi-source research with citations. The README says the public endpoint needs no account or API key and is rate-limited per IP.

• Trust Score 85, security 89 (Review), only info-level findings
• 1 GitHub star; not owner verified on HOL; MCP Hardened badge
• What Grok loads: 2 skills and one hosted MCP server at api.keenable.ai/mcp. Nothing runs on your machine.

The README says to open /plugin in Grok Build, search for Keenable and install it. Keenable isn't in xAI's official marketplace, so if the search comes up empty, the GitHub form from xAI's docs works too:

grok plugin install keenableai/keenable-grok-plugin --trust

Example prompts from the README:

Search for the latest changes to the Model Context Protocol and cite the primary sources.
Research the leading agent observability platforms and compare them with citations.

Pros: the quickest pick to try, with no key and no local process. Higher limits are available with a Keenable key sent in a header, which the plugin itself never sends.

Cons: every search you run goes to Keenable. The repo has one star, the owner hasn't verified it on HOL, and the README's install steps assume a marketplace listing it doesn't have.

HOL listing · GitHub

6. Cargo: lead lists and CRM updates from a prompt

Cargo gives the agent go-to-market jobs on top of the Cargo CLI: build lead and account lists, find and verify work emails, enrich companies from licensed data providers, score leads, sync a CRM such as HubSpot or Salesforce, and watch signals like job changes and funding. Two agents come with it, one that plans a costed run and one that sources lists in parallel. New accounts get 100 free credits for 30 days, per the manifest.

• Trust Score 88, security 95 (Safe), only info-level findings
• 22 GitHub stars; not owner verified on HOL; Scanner Clean badge
• What Grok loads: 19 skill folders named in the manifest, 2 agents, and one hook that runs before every shell command

Install, from the README. Grok plugins don't install programs, so the CLI is a separate step:

grok plugin marketplace add getcargohq/cargo-skills
grok plugin install cargo --trust
grok plugin enable cargo
npm install -g @cargo-ai/cli
cargo-ai login --oauth

For Grok Bot, the README says to open Connect apps, then Team plugins, then Cargo, then Connect. Example prompts from the README:

Find me 50 VPs of Sales at Series B fintechs and verify their emails.
Enrich this CSV with company size, funding stage, and tech stack.

Pros: it documents both Grok Build and Grok Bot. The README says the Grok package doesn't run Claude-only session hooks, install software on its own, send session data or embed a token.

Cons: that one hook exists to approve Cargo commands for you. The README says reads, queries and run operations go through without a prompt, while logins, token minting, deploys and deletes still ask. If you want to see every command, leave this plugin's hook off. Runs cost credits.

HOL listing · GitHub

7. Homie: make a game and put it online

Homie makes multiplayer web games, music and video, then publishes them from a studio site on your own Cloudflare account. The README says the free Cloudflare plan is enough to start. Its skills cover planning, game building, sound, music, art, video, playtesting and publishing, and a hosted server at homie.rocks/mcp supplies the tools.

• Trust Score 84, security 96 (Safe), only info-level findings
• 7 GitHub stars; owner verified on HOL; Scanner Clean badge
• What Grok loads: 24 skills, the hosted MCP server, and hooks before and after tool calls and when you send a prompt

Install, from the README (it says to trust the plugin when asked):

grok plugin install homie-rocks/homie#plugins/homie

The README also says Grok Bot installs Homie itself when told to read homie.rocks/install.md, and any MCP client, Grok chat included, can add https://homie.rocks/mcp as a custom connector. Example prompt from the README:

Make a game for my pub and put it online.

Pros: the hooks guard you rather than skip prompts. A production deploy, a change to the studio's Cloudflare account, or a paid call over budget waits until you type proceed with a code. The README explains how a missing hooks entry once let a deploy run without that check in Grok, and how it was fixed.

Cons: the hooks run on every edit, shell command and MCP call, so this plugin is active the whole session. Publishing needs a Cloudflare account, and the README offers only one example prompt.

HOL listing · GitHub

8. Second Claude Code: research and writing that doesn't skip the check

Despite the name, Second Claude Code installs in Grok with its own Grok manifest. It is a knowledge-work plugin built around one loop: gather and analyze a brief, draft from the approved plan, check the draft with a panel of 2 to 5 reviewers, then cut or send gaps back. Separate commands handle research, writing, review, refinement and coaching.

• Trust Score 87, security 89 (Review), 0 high findings
• Findings behind the score: two low Dependabot notes
• 41 GitHub stars, 9 forks; not owner verified on HOL
• What Grok loads: the README counts 16 skills, 19 commands, 17 agents and 10 hook events. Its .mcp.json starts one local server for workflow state and ships two more, Playwright and MMBridge, switched off.

Install, from the README:

grok plugin install unclejobs-ai/second-claude-code --trust

Example commands from the README:

/scc:research "AI agent frameworks" --depth medium
/scc:review draft.md --preset content

Pros: the review step is built in, so a draft gets checked before it comes back to you. The optional browser and bridge servers stay off unless you turn them on.

Cons: the most moving parts of any pick, with hooks on session start, every prompt, subagent start and stop, and context compaction, all running Node scripts. The README says Grok doesn't mirror every Claude Code slash command and that you ask for skills by name, so the examples above may need rewording in Grok.

HOL listing · GitHub

9. telepathy: let your agent sessions talk to each other

telepathy lets Grok, Claude Code, Codex, Cursor and other agent sessions on your machine find each other, send messages and hand off work. An idle session can wake up when a message arrives. An optional bridge extends this to your other devices, and the README says messages stay local unless you start that bridge or connect a Grok Bot.

• Trust Score 77, security 85 (Review), 0 high findings
• Findings behind the score: low ones for a missing SECURITY.md and README in the scanned folder. HOL takes 10 percent off for no scanner in CI.
• 5 GitHub stars; owner verified on HOL; launched on HOL Plugins in September
• What Grok loads: 1 skill, a local MCP server started with Node, and hooks on session start, after each tool call and at the end of a turn

Install, from the README:

grok plugin install Winterrks/telepathy#plugin --trust

For Grok Bot, the README says telepathy grokbot connect joins a Bot routine as a session. Example prompts from the README:

Ask the Codex session what it's working on.
Ask the Codex on my desktop whether the training run finished.

Pros: two agents in one repo can split work instead of colliding. The README says there is no telemetry and no server of theirs in the middle.

Cons: it needs Node.js 20 or newer, and its hooks run after every tool call so messages can reach the session. HOL scores its maintenance at 10 out of 100. Messages from other agents land in your session, so the hooks are what let them in.

HOL listing · GitHub

10. Alcove: search your own project docs

Alcove keeps your private project docs in one place and lets the agent search them with a mix of keyword and meaning-based search, so it stops asking you to explain the same architecture every session. It also indexes code and can lint docs against team standards. It runs locally.

• Trust Score 79, security 92 (Safe), only info-level findings
• 16 GitHub stars; not owner verified on HOL; Scanner Clean badge. HOL takes 10 percent off for no scanner in CI.
• What Grok loads: 1 skill and two hooks. At session start a Node script installs or updates the alcove program. At session end a second hook runs alcove reap.

Install, from the README, then run setup once:

grok plugin marketplace add epicsagas/plugins
grok plugin install alcove@epicsagas --trust
alcove setup

Example prompts from the README:

/alcove how is auth implemented?
/alcove search auth flow

Pros: your docs stay on your machine, and the epicsagas marketplace pins Alcove to the same commit HOL scanned.

Cons: xAI's docs say marketplaces deliver files, not programs, but this plugin's session-start hook installs one. We read the script: it tries Homebrew on macOS, then cargo binstall, then downloads and runs the installer from the latest Alcove release on GitHub. The plugin is pinned, but the program it installs is not. That is the most any pick here does on its own, which is why it comes last.

HOL listing · GitHub

What we left out

• pstack (4,171 stars), the most starred Grok listing. Its README has no Grok install. xAI's marketplace lists a pstack from a different repo.
• agmsg and AgentCall: 47 and 79 high findings.
• Concord (1,005 stars): its Grok setup lives in a separate doc, and HOL's file list shows no Grok manifest for it.
• HyperGrok: a Hyperliquid trading desk built mainly for Grok Bot, with five medium findings and a Trust Score of 71.
• Nimble, Equibles and NeverApply: good hosted-data plugins, but Keenable covers web data with no account, Equibles documents only a local-folder Grok install, and NeverApply gives no Grok Build command.
• OpenSea Skills, Grok Imagine Cinematic Studio, Alloy and MetaMask Agent Wallet: one or more high findings.
• HOL Guard's own Grok plugin, left out because it is our product.

If you want the catalog ranked by score rather than by what each plugin turns on, the Best Grok plugins page does that.

Build your own Grok plugin

A Grok plugin can be a folder with nothing but skills/ in it. Add .grok-plugin/plugin.json when you need a name, a version, or paths that differ from the standard folders. Several picks here name their hooks file in the manifest for a reason: Homie's README explains that Grok loaded no hooks at all from its plugin until the manifest pointed at hooks/grok.json. If you already ship a Claude Code plugin, Grok will read it, but a native manifest lets you give Grok its own hooks and MCP settings, which is what Cargo, Homie and telepathy do. In hooks, use GROK_PLUGIN_ROOT for the install folder and GROK_PLUGIN_DATA for state.

To share it, list the plugin in .grok-plugin/marketplace.json in a git repo and pin each remote plugin to a full commit SHA, since teams can require pinned installs. Run grok plugin validate before you publish. xAI takes submissions to its official marketplace as pull requests to xai-org/plugin-marketplace.

Before you publish, run the HOL Plugin Scanner on the repo. It can run in GitHub Actions, which also avoids the 10 percent Trust Score cut that six picks here took. Then submit the plugin to HOL Plugins for a public listing. To review what a trusted plugin does while it runs, HOL Guard checks agent actions before they happen.

What do people ask about Grok plugins?

What is the best Grok plugin in 2026?

It depends on the job and how much you want to trust. Keenable is the quickest to try because it needs no account. go-ultimate has the highest Trust Score of the ten at 92. Antidote and Dryforge are skills only, so trusting them turns on nothing that runs. For the full catalog ranked by Trust Score, see the Best Grok plugins page on hol.org.

How do I install a plugin in Grok Build?

Run grok plugin install with a GitHub owner/repo, a git URL or a local folder, and add --trust. To use a marketplace, run grok plugin marketplace add first, then install by name. Inside Grok Build, type /plugins or /marketplace and press i on a plugin.

What does --trust do in Grok Build?

It confirms the install and lets the plugin load its skills, commands, hooks, MCP servers and LSP servers. Without it, Grok shows the source, warns you and stops. Plugins placed in ~/.grok/plugins/ are trusted automatically, while project plugins in .grok/plugins/ need trust.

Do Claude Code plugins work in Grok?

Yes. xAI's docs say Grok reads Claude Code marketplaces, plugins, skills, MCP servers, agents and hooks with no extra setup, and Grok sets the CLAUDE_PLUGIN_ROOT variable for plugin hooks as well as its own. That is why 44 of HOL's 68 Grok listings ship a Claude Code manifest.

Can I use these plugins in Grok Bot or Grok chat?

Not as Grok Build plugins. Grok Bot adds apps from Connect apps, and Grok chat adds connectors at grok.com/connectors, including custom MCP servers. Plugins with a hosted MCP server, such as Homie, can be added that way, and Cargo's README says it is listed under Team plugins in Grok Bot.

What is the difference between a Grok skill and a Grok plugin?

A skill is a SKILL.md file with instructions for one kind of task. A plugin is a package that can bundle skills with commands, agents, hooks and MCP servers, so it can give Grok live tools and run code on events as well as give instructions.

Are Grok plugins safe to install?

Not automatically. xAI's docs say plugins run with your privileges. The official marketplace pins each remote plugin to a commit, but most plugins in the HOL catalog install straight from GitHub. Read the manifest, check for hooks and MCP servers, look at the HOL listing's label and findings, and remember that HOL's scan was partial for all ten picks.