Claude Code plugins package reusable workflows and external tools for Anthropic's coding agent. Vercel, GitHub, Sentry, Linear, Figma, and Hugging Face can supply the deployment knowledge, repository context, or service access a development task needs beyond the files in your checkout.
The packages here target Claude Code. Claude chat connectors and browser extensions have different installation paths and capabilities. For a team choosing plugins, the useful questions are which task each package supports, where it comes from, and which accounts it needs to reach.
Copy link to headingUnderstand what a Claude Code plugin installs
Claude Code loads a plugin's components together. Skills provide task instructions; an MCP server exposes tools and data the agent can request. Plugins can also include subagents and hooks for more specialized behavior.
The marketplace is a distribution mechanism. Anthropic's official catalog and vendor-maintained catalogs can provide packages with different components. Check the package details before assuming it contains both skills and an MCP server.
Reusable skills and MCP configuration are also the focus of Agent Plugins. Client-specific hooks and installation behavior require separate checks, even when two clients use instructions from the same repository.
Copy link to headingInstall plugins with an explicit source and scope
With Claude Code installed, start an interactive session from your project directory:
claudeOpen the plugin browser inside that session:
/pluginThe Discover tab lists packages from configured marketplaces. Anthropic's official marketplace is added when you first start an interactive terminal session. Vendor catalogs can be registered separately, as shown in the Hugging Face section below.
For a named package, use the in-session command to open its details and choose an installation scope:
/plugin install vercel@claude-plugins-officialUser scope applies across projects on your machine. Project scope records the enabled plugin in the repository's shared settings; local scope limits it to your use in that repository. Read the installation result to see whether the package is active or needs a reload or server configuration.
You can also install directly from your shell. The individual sections below use that form, which defaults to user scope. Add --scope project when you intend to record a shared project installation. To inspect installed packages from the terminal, run:
claude plugin listUse Installed in the plugin browser to manage packages and Errors to investigate loading failures. Inside a session, the MCP management command shows server status and authentication options:
/mcpCopy link to headingSelect plugins by development task
These six packages cover different kinds of application work. Vercel's own plugin is included for platform guidance. GitHub and Linear primarily package service access; Sentry and Figma add workflow instructions, and Hugging Face offers a CLI skill through its vendor marketplace.
Copy link to heading1. Vercel for platform guidance
The Vercel Plugin gives Claude Code skills for Vercel development and a connection to Vercel's MCP server. It can guide deployment work and AI SDK integration, with specialist agents for architecture, deployment, and performance tasks.
Have Node.js 18 or later and Bun installed. Then add the package from your terminal:
claude plugin install vercel@claude-plugins-officialFor a project moving toward its first preview, try:
Review this application's Vercel setup. Identify the build command and required environment variables, explain any configuration gaps, and prepare the smallest changes needed for a preview deployment.
The bundled Vercel connection requires authentication before Claude Code can inspect your projects or deployment logs. Confirm the team and project it reaches before asking it to deploy.
Default hooks keep automatic context loading limited to empty directories and detected Vercel, Next.js, or eve projects. You can request skills explicitly when investigating a specific concern. Installing the plugin doesn't mean every tool call triggers a new set of platform instructions.
Copy link to heading2. GitHub for repository and review workflows
The GitHub package in Anthropic's catalog connects Claude Code to GitHub's remote MCP server. Use it when issue discussions, review comments, or Actions results should inform a change to local code.
Install from your shell:
claude plugin install github@claude-plugins-officialThis package's MCP configuration expects a personal access token in the environment that starts Claude Code. Provide it under this variable name using your usual secret-management process:
GITHUB_PERSONAL_ACCESS_TOKENGitHub token permissions determine which repositories and operations the connection can use. Keep the token out of shared plugin files and repository settings.
Give Claude Code a review task with a defined output:
Read this pull request's unresolved comments. Check each one against the current code, implement the changes that still apply, and draft replies with test evidence for my review.
Ask for the test command and result alongside each proposed reply. The review can then show which comments led to verified code changes and which still need discussion before you publish an update.
Copy link to heading3. Sentry for setup and production debugging
Sentry's Claude Code plugin contains setup and debugging skills alongside the hosted Sentry MCP server. It can guide SDK instrumentation, investigate issues, and help configure monitoring for the application.
Use Sentry's installable distribution through Anthropic's catalog:
claude plugin install sentry@claude-plugins-officialAfter installation, use the MCP connection flow to sign in to Sentry. The hosted server uses OAuth; self-hosted installations need the separate Sentry CLI connection path.
For a release regression, ask:
Investigate the new errors in this Sentry project after the latest release. Group the relevant events, trace one reproducible failure to local code, and prepare a fix with a regression test.
The skill can request Seer analysis where the account supports it, but the investigation still needs usable telemetry. Missing source maps or release information can leave the cause uncertain. Have Claude Code distinguish what the event establishes from what the code suggests before it changes issue status.
Copy link to heading4. Linear for turning requirements into work
The Linear plugin configures Linear's hosted MCP server for issue and project operations. It brings the team's recorded requirements into the coding session without requiring you to copy the whole ticket discussion.
Install it in your terminal:
claude plugin install linear@claude-plugins-officialThen use /mcp inside Claude Code to complete Linear's authentication flow. Check the workspace before requesting an issue or project change.
For implementation planning, try:
Compare this Linear issue's acceptance criteria with the code and tests on this branch. List the remaining work, implement the agreed behavior, and draft a progress comment that identifies any unresolved requirement.
Separate a completed code change from a completed ticket. The acceptance criteria may also require a design review or a deployed behavior check. Linear's connection can update the record when those conditions are met; a read-only connection can provide the same requirements context without issue-editing access.
Copy link to heading5. Figma for designs that reuse your components
The Figma plugin includes its MCP configuration and skills for design implementation and Code Connect. Claude Code can use a frame's structured layout and variables while editing the corresponding application code.
Install the package from your terminal:
claude plugin install figma@claude-plugins-officialAuthenticate the connection, then supply a frame link with the implementation constraints:
Implement this billing page with our existing layout and form components. Use mapped components where available, flag any design state without an equivalent, and verify keyboard navigation and responsive behavior.
Code Connect supplies relationships between Figma components and code. Those mappings help Claude Code find the right implementation, while your prompt establishes requirements that a static frame may not express.
Figma's remote server also supports canvas workflows, subject to the available tools. File access and read limits depend on the connected account, plan, and seat. Some image-generation and Figma Make actions consume AI credits, so scope those separately from reading a design.
Copy link to heading6. Hugging Face for Hub operations and model workflows
Hugging Face's CLI skill plugin teaches Claude Code to use the hf CLI for model and dataset work, repository operations, and remote Jobs. This path installs the hf-cli package from Hugging Face's own catalog.
Install the Hugging Face CLI first using its setup instructions. Then run these commands inside a Claude Code session:
/plugin marketplace add huggingface/skills/plugin install hf-cli@huggingface-skillsThe repository name registers the catalog; the name after @ identifies that catalog during installation. This skill-based package doesn't by itself configure an MCP connection. Hugging Face offers MCP access separately if that fits your workflow.
For an application adding semantic search, try:
Inspect these embedding model cards and licenses. Propose an evaluation dataset from our representative queries, estimate the resources needed, and wait for approval before launching a remote job.
Authenticate the CLI for private or write operations. Additional skills can supply more specialized workflows, while Jobs billing remains separate from plugin installation. Decide the evaluation and resource limits before starting compute.
Copy link to headingUse several plugins to finish a project task
For a production bug, ask Claude Code to read the Linear issue and the relevant Sentry events before changing files. It can compare the reported behavior with the local implementation, identify a failing case, and prepare a patch. Include the repository's test commands and require an explanation if the failure cannot be reproduced.
GitHub can add review discussion and CI context to that investigation. Once the patch passes its checks, use Vercel guidance to prepare a preview and verify the affected behavior there. The final issue update should distinguish the code change, test result, and deployment state so teammates can see what is complete.
Keep external actions explicit in the request. Claude Code can prepare a proposed comment without posting it, and it can inspect deployment requirements before creating a deployment. Which actions it can perform depends on the installed package and connected account.
Copy link to headingRoute Claude Code through Vercel AI Gateway
AI Gateway can route Claude Code's model requests and bring them into Vercel's usage and spend views. The Vercel Plugin's ai-gateway skill has a separate development role: helping you integrate Gateway into the application, including routing and fallback configuration. The plugin doesn't establish the coding agent's provider connection.
Configure Claude Code from your terminal:
pnpm dlx vercel@latest ai-gateway setup --agent claude-codeThe command configures API-key authentication and model discovery, and backs up the files it changes. Add --dry-run to inspect the proposed setup first. Start a new terminal session so shell changes load, run Claude Code, and use /model to select a Gateway model. Confirm a request in the AI Gateway Overview.
This setup uses API-key billing. The Claude subscription route has separate manual configuration and draws on your subscription limits. The CLI setup command doesn't configure that route. Choose the authentication path that matches how you intend to pay for model access.
Copy link to headingPackage and distribute a team workflow
Turn a repeated review procedure into a plugin when several repositories need the same instructions. For example, a release-readiness skill could check that the pull request has test evidence and a working preview, then draft the team's release note format.
For a native Claude Code package, put the manifest in .claude-plugin/plugin.json and keep components such as skills/ beside that directory. The plugin authoring guide provides the full layout. Validate and try a local package from your terminal, replacing the example path with your plugin directory:
claude plugin validate ./release-reviewclaude --plugin-dir ./release-reviewThe directory flag loads the package for that session. For regular team installation, distribute it through a marketplace repository and give teammates the catalog and package names.
Project scope can commit the enabled-plugin setting for collaborators, but it doesn't download the package onto their machines. Each collaborator still installs it for their clone. Keep service credentials outside the repository so shared settings express the team's tooling choices without sharing a person's access token.
If the workflow also needs to run in other agents, use the Agent Plugins guide to package portable skills and MCP configuration. Check each client's support before carrying over Claude Code hooks or commands.
Copy link to headingFrequently asked questions
Copy link to headingWhere do I find plugins for Claude Code?
Run /plugin in an interactive Claude Code session to browse the marketplaces you have configured. Anthropic's official catalog is added during the first interactive terminal session, and you can register vendor or team repositories separately. Use the package name and marketplace name together to identify the installation source.
Copy link to headingDoes installing a plugin connect all my accounts?
No. Plugins can provide connection configuration, but each service still requires its own authentication and permissions. Some use an interactive sign-in; the GitHub package shown here expects a personal access token in the environment. Check server status with /mcp inside Claude Code.
Copy link to headingWhich installation scope should a team use?
Project scope records the enabled plugin in shared repository settings. Collaborators still need to install the package for their own clone. Choose local scope for a personal addition limited to that repository, or user scope for your projects across the machine.
Copy link to headingDo these instructions also install Claude chat connectors?
No. These commands install packages for Claude Code's coding sessions. Claude chat connectors have their own setup, and a service with both integrations can expose different capabilities through each one.
Copy link to headingCan I keep a Claude subscription while using AI Gateway?
Yes. Claude Code can authenticate with a Claude subscription while routing requests through AI Gateway and using the subscription's limits. Configure that path separately from the CLI's API-key setup, then verify requests in Gateway. Installing the Vercel Plugin alone doesn't select either billing route.