Claude Code Templates: an npx installer for 1,900 Claude Code components, after the 1.31.0 flag cull

One npx call can stack an agent, a command and an MCP integration

The project is two things that share a name: the catalog at aitmpl.com and the claude-code-templates npm package that installs from it. The catalog holds about 1,900 components across skills, agents, slash commands, MCP integrations, settings, hooks, mods, loops, sandboxes and project templates. On the site you can search, preview every file a component writes, and combine several components into one install command.

The CLI takes one flag per component type, and flags can be combined in a single run. The headline example installs a frontend agent, a test-generation command and the GitHub MCP integration together:

npx claude-code-templates@latest --agent development-team/frontend-developer --command testing/generate-tests --mcp integration/github-integration --yes

Running it with no flags starts an interactive browser instead:

npx claude-code-templates@latest

Component names follow a category/name pattern, such as creative-design/frontend-design for a skill or mcp/mcp-timeouts for a setting:

npx claude-code-templates@latest --skill creative-design/frontend-design --yes
npx claude-code-templates@latest --agent development-tools/code-reviewer --yes
npx claude-code-templates@latest --command performance/optimize-bundle-size --yes
npx claude-code-templates@latest --setting mcp/mcp-timeouts

Most of these examples end in --yes, which turns them into one-line installs you can paste into a script, so decide what you are installing on the site first. package.json registers two binary names for the same entry point, claude-code-templates and a short cct, both pointing at cli-tool/bin/create-claude-config.js.

Ten component types with very different blast radius

The catalog is uneven by design. Skills are the largest group at 890+, described as reusable capabilities with progressive disclosure, with examples such as PDF processing, frontend design and code review. Agents number 420+ and are domain specialists like a security auditor, frontend developer or database architect. Commands, at 280+, are custom slash commands such as /generate-tests, /optimize-bundle-size and /security-audit. MCPs (100+) connect external services including GitHub, PostgreSQL, Stripe, Supabase and Sentry.

The smaller groups touch Claude Code itself. Settings (75+) cover timeouts, permissions and status lines. Hooks (60+) are automation triggers, for example enforcing conventional commits or blocking direct pushes. Mods (45+) are plugins that change Claude Code's own interface and behavior, from secret redaction and protected paths to terminal games, and they require Claude Code 2.1.287 or newer. Loops (18) are autonomous workflows defined by a goal, an interval and a stop condition, such as a docs sweep or build-test-fix. Templates (14) are complete project configurations for languages and frameworks.

A skill or agent mostly adds instructions. A hook, setting or mod changes what Claude Code does on its own, so those are the ones to read before installing. Current counts live in dashboard/public/counts.json, so the numbers above move as contributions land.

Version 1.31.0 removed seventeen flags at once

The repository's package.json reads version 1.31.0, and that release cut the CLI down hard. The removed flags are --plugins, --skills-manager, --command-stats, --hook-stats, --mcp-stats, --agents, --chats-mobile, --teams, --2025, --studio, --sandbox, --tunnel, --create-agent, --list-agents, --update-agent, --remove-agent and --clone-session. Running any of them now prints what replaced it, which helps interactively but still breaks a script or a tutorial that calls them.

The replacement for listing what is installed is no longer a flag at all. It is a mod: install the stack viewer, then type /stack inside Claude Code to see plugins, skills, commands, hooks and MCPs per scope. The install line is npx claude-code-templates@latest --mod ui/stack-viewer. That route needs mod support, which ties it to Claude Code 2.1.287 or newer.

There is also a mismatch to be aware of when reading release notes. The most recent GitHub releases are v1.29.5 on September 9, 2026, which added installing TypeScript function hooks from the catalog, v1.29.6 on September 17, which made --mod install Anthropic-layout plugins with 24 mods, and v1.29.7 on October 9. None of those three is the 1.31.0 that package.json names, and the removals are detailed in CHANGELOG.md, so read the changelog before upgrading a setup that scripts the CLI.

Sandboxes stayed in the catalog after the --sandbox flag went away

One casualty of 1.31.0 is easy to misread. The component table still lists eleven sandboxes, launchers that run Claude Code in an isolated environment, with E2B, Cloudflare and Docker as examples. The CLI's --sandbox flag that used to start them was removed in the same release.

The sandbox launchers are kept as files in cli-tool/components/sandbox/ in the repository. The consequence is that sandboxes are now something you take from the source tree and run yourself, not something the installer starts for you. Anyone who followed an older walkthrough that used npx with --sandbox will get a message pointing elsewhere instead of a running environment, and needs to open that directory and follow the launcher for the provider they use.

The analytics and chat dashboards bind to 127.0.0.1 because they serve your history

Beyond installing components, the CLI carries three local tools. Two of them are web dashboards, and both listen on 127.0.0.1 only and answer only to localhost, because they serve your Claude Code conversation history.

The analytics dashboard monitors sessions in real time with live state detection and usage metrics on http://localhost:3333:

npx claude-code-templates@latest --analytics

The conversation browser lets you read and search past conversations in a mobile-friendly UI on http://localhost:9876:

npx claude-code-templates@latest --chats

Mobile-friendly does not mean reachable from your phone by default. The public --tunnel option was removed in 1.31.0, and the supported way to open the browser from another device is SSH port forwarding, ssh -L 9876:localhost:9876 followed by your host. If you relied on the tunnel for quick remote access, that path is gone, and you need SSH access to the machine running Claude Code.

The third tool is a health check for your Claude Code installation and project setup. It answers to --health-check, and --health, --check and --verify are aliases:

npx claude-code-templates@latest --health-check

A community catalog means installing other people's prompts and hooks

The catalog is assembled from many sources. The attribution section credits outside collections, for example 139 scientific skills for biology, chemistry, medicine and computational research from K-Dense-AI/claude-scientific-skills under the MIT License. Contributors add skills, agents, commands, MCPs, settings, hooks and mods through pull requests, and every pull request is checked by Claude Code's plugin validator and scanned by NVIDIA SkillSpector.

Those checks are a floor, not a review of whether a component suits your project. A hook that blocks direct pushes or a mod that redacts secrets is doing real work inside your Claude Code session, and an agent's prompt shapes what Claude Code does with your files. The aitmpl.com preview that shows every file a component writes is the practical safeguard here. Use it before running an install with --yes, particularly for hooks, settings and mods, which act without being asked.

The repository's .env.example configures the project's own services

Cloning the repository can give the wrong impression about setup. The root .env.example asks for a long list of keys: RAPIDAPI_KEY, SERPAPI_KEY and SERPER_API_KEY for job search APIs, GITHUB_TOKEN, GOOGLE_API_KEY for blog image generation, SUPABASE_URL and SUPABASE_API_KEY for download statistics, NEON_DATABASE_URL for Claude Code changelog monitoring, Discord bot and webhook values for stats and changelog posts, COMPONENTS_API_URL set to https://aitmpl.com/components.json, and SENTRY_DSN for the dashboard API routes.

None of those keys appear in the npx commands above. They belong to the site, the dashboard and the workers in the same repository, which also holds cli-rust/, cloudflare-workers/, dashboard/ and database/ next to cli-tool/. Two details matter to contributors. SUPABASE_API_KEY must be the service_role key, because row level security is on with no policies and the anon key reads nothing. And in production SENTRY_DSN is set as a Cloudflare Pages secret with wrangler, not through this file. The published npm package ships only cli-tool/bin/, cli-tool/src/, cli-tool/templates/, cli-tool/package.json and the README.

Editorial conclusion

Claude Code Templates suits Claude Code users who want ready-made agents, skills, slash commands, hooks and MCP integrations installed with one npx command, plus local analytics and a conversation browser. It does not suit anyone who relies on the flags removed in 1.31.0, such as --sandbox, --tunnel or --agents, or who wants a vetted vendor library rather than a community catalog. Before installing a hook or mod, preview every file it writes on aitmpl.com, and check that your Claude Code is 2.1.287 or newer before using mods.

Frequently asked questions

Official sources

1. davila7/claude-code-templates on GitHub
2. License: MIT
3. Project website
4. README
5. Releases