The AI coding assistant space is evolving fast. Both Anthropic and Google are building ways to customize their CLI tools. In the same week, Google launched Gemini CLI Extensions and Anthropic launched Claude Code Plugins.

Let’s look at how they compare.

The Big Picture

Claude Code has plugins. Gemini CLI has extensions. Different names, similar goals. Both let you package up customizations and share them with your team or the world.

Think of them as app stores for your AI coding assistant. You install what you need. You disable what you don’t.

What They Have in Common

The overlap is actually pretty significant.

Both support MCP servers. That’s the Model Context Protocol, which connects your AI to external tools and data sources. Need to pull from an API? Query a database? MCP handles it.

Both let you create custom commands. In Claude Code, these are slash commands. In Gemini CLI, they’re also slash commands but stored as TOML files. Same idea either way.

Both support adding context. Claude Code uses hooks and system prompts. Gemini CLI uses context files. The mechanics differ but the purpose is the same. Give the AI more information about your project.

Installation works similarly too. You can pull from GitHub or point to a local directory. Both have update commands. Both let you enable and disable things.

Where They Differ

The details matter though.

Claude Code calls its customization points “hooks.” These let you modify behavior at specific points in the workflow. Gemini CLI doesn’t have hooks. Instead, it has tool exclusion lists. You can block specific tools or commands from running.

Claude Code has this marketplace concept. Anyone can host a marketplace. Just need a git repo with the right file structure. It’s decentralized. Gemini CLI doesn’t mention marketplaces in their docs.

Command conflicts get handled differently. Say you install an extension that adds a /deploy command. But you already have a /deploy command. Gemini CLI has clear rules. User commands win. Project commands come next. Extension commands are last. If there’s a conflict, the extension command gets renamed with a prefix. So /deploy stays yours and /gcp.deploy is the extension version.

Claude Code doesn’t spell out these rules in the provided docs.

Gemini CLI supports variable substitution. You can use ${extensionPath} or ${workspacePath} in your config files. Handy for when you need to reference directories. Claude Code docs don’t mention this.

Why This Matters

Both approaches work. The question is what fits your workflow.

Claude Code’s marketplace model is interesting. It’s more community-driven. Anyone can publish. Anyone can curate. You might see more experimentation here.

Gemini CLI’s conflict resolution is clearer. If you’re worried about namespace collisions, that matters. The variable substitution is nice too. Small feature but it makes config files cleaner.

Both systems are still new. Claude Code plugins just hit public beta in October 2024. Gemini CLI extensions are at v0.4.0. Expect things to change.

The Takeaway

We’re seeing a pattern emerge. AI coding assistants need to be extensible. Monolithic tools don’t cut it. You need customization. You need to share that customization.

Claude Code and Gemini CLI are taking slightly different paths to the same destination. Both are betting on MCP. Both are betting on custom commands. Both are betting on community contributions.

Let’s see how this evolves over next few months.