Gemini CLI extensions launched on October 8, 2025. One week later, there were 106 extensions available in the marketplace. The top extension, context7, already had 34,461 downloads. GitHub’s official integration had 23,767.
For context, the Model Context Protocol ecosystem took months to reach 1,000 servers after Anthropic introduced it in November 2024. Gemini CLI hit over 1 million users in its first three months. The extension marketplace is moving fast.
I wanted to understand what this early data tells us about what developers actually need from AI tooling. The numbers reveal some interesting patterns.
Two Types of Extensions Living Side by Side
In a previous article, I covered how MCP servers connect AI to data and tools while Claude Skills provide task-specific intelligence. Different approaches for different problems. The Gemini CLI marketplace shows both living in the same ecosystem.
The breakdown is roughly 85% MCP extensions and 40% Context extensions, with some overlap. MCP extensions connect to external systems like databases, GitHub, Stripe, and Figma. Context extensions are custom commands, prompts, and workflows.
This is not an either/or situation. Developers need both. That matters.
Connection Won the First Week
Looking at what developers installed first tells you what they needed most urgently. The top four extensions are:
- context7 (34,461 downloads) - up-to-date documentation
- GitHub (23,767 downloads) - official integration
- Chrome DevTools (11,714 downloads) - browser automation
- Database Toolbox (11,018 downloads) - access to 30+ datasources
Notice the pattern. Every single one provides access to information and systems.
The marketplace has over 30 database extensions alone. PostgreSQL, MySQL, BigQuery, Neo4j, Redis, MongoDB, Pinecone, Elasticsearch. The list goes on. Every team has different data sources. No single standard is winning yet. Maybe no standard should win.
You cannot optimize workflows until you can access your data. Developers installed connection tools first. Task-specific intelligence comes second.
The Long Tail Problem
Official vendor extensions are succeeding. GitHub, Stripe, Shopify, Dynatrace. These have thousands of downloads. Meanwhile, more than 60 extensions have fewer than 50 downloads each.
This is completely normal for week one. Every app store looked like this in the early days. Lots of experiments. Some winners. Lots of noise.
The difference here is the ranking system. Downloads, GitHub stars, community feedback. Quality signals take time to emerge. Give it six months. The useful extensions will rise to the top.
Compare this to the MCP ecosystem I wrote about previously. That was the wild west. Some servers were just node apps wrapping CLI tools. Others were remotely served. Authentication was a mess. It still kind of is.
Google’s approach with Gemini CLI is different. Open marketplace with ranking signals instead of curated gatekeeping. I think this will work better long term. But it requires patience in the short term.
Why Versatility Matters
Having both MCP and Context extensions in the same marketplace is actually the point.
MCP extensions solve the problem of connecting to your stuff. GitHub integration for code context. Database connections for data access. API integrations for third-party services.
Context extensions solve the problem of teaching it your process. Code review workflows. Documentation templates. Project-specific commands.
Here’s a real example from my own workflow. I use an MCP extension to connect to BigQuery for analytics data. I use a Context extension with custom commands for generating weekly reports in our specific format. Together, the AI can access the data and knows how we want it presented.
One-size-fits-all does not work. Different problems need different tools.
The Gaps Tell a Story
What’s missing from the marketplace right now is as interesting as what’s there.
Testing frameworks are barely present. Logging and monitoring have thin coverage outside of Dynatrace. CI/CD integrations are underrepresented. Documentation generators are surprisingly absent, which is ironic given that context7 dominates the rankings.
These gaps reveal where opportunity still exists. Week one adoption focused on data access and system integration. The workflow optimization tools will come next.
What’s Working and What’s Still in Flight
The things working well in week one:
Low barrier to entry means anyone can publish an extension. MCP standardization builds on an existing ecosystem. The marketplace supports both connection and workflow tools. A ranking system is in place for quality signals to emerge. Major vendors are participating, which provides trust and distribution.
The things still in flight:
Discovery is hard when you have 106 options and counting. Quality varies wildly, which you expect in week one but still makes choices difficult. Token bloat from MCP servers remains an issue from the broader ecosystem. There is no clear guidance yet on which extensions to start with for common use cases.
None of these are permanent problems. They are early-stage problems.
The ranking data will mature over the next three to six months. The community will coalesce around best-in-class extensions for each category. Google may add a curation layer like featured extensions. We might see extension bundles for common workflows. Better token management patterns will emerge as people figure out what actually works.
This is what healthy ecosystem growth looks like.
Where to Start Today
If you want to try Gemini CLI extensions now, here is what I would recommend.
Start with official vendor extensions for tools you already use. GitHub if you write code. Stripe if you handle payments. These have the trust and maintenance you want.
Try context7 if you work with multiple frameworks. Keeping documentation current is harder than it sounds. Having that automated saves real time.
Pick one database extension for your primary data source. Don’t install five different database tools right away. Figure out what you actually need first.
Wait on the long tail unless you have a very specific need. Let the ranking system do its work. The marketplace will look very different in six months.
The Cambrian Explosion Continues
I wrote previously about how we are in an AI Cambrian Explosion with specialized models, RAG frameworks, and autonomous agents emerging at unprecedented pace. The challenge is connecting AI models to the data, tools, and workflows they need to be useful.
This marketplace data shows that challenge playing out in real time. Developers need both connection and customization. They installed connection tools first. Task-specific intelligence is already following.
Early chaos is normal. Quality signals take time. The open approach will likely win over gatekeeping in the long run.
We are watching the ecosystem figure out what it needs. Week one data shows developers need access to their systems first. The workflow optimization will come. The gaps will fill. The long tail will sort itself out.
Give it time. This is just getting started.