
Introduction We had a few big announcements from Google I/O 2026 in the AI space. One was...
We had a few big announcements from Google I/O 2026 in the AI space. One was the announcement of Gemini Flash 3.5, which went straight to general availability (GA). Another was the launch of the Google Antigravity (aka “Agy”) ecosystem.

The Agy suite supports and enables Gemini 3.5 from day 1 and is made up of a number of products:
google.antigravity we can programmatically leverage Antigravity’s capabilities.
I can understand Google’s rationale for bringing Gemini CLI into the Antigravity suite. Previously these products were developed independently and in parallel. As a result they had different standards and approaches for configuration (such as MCP servers), and where to put things… like skills. This creates a certain amount of friction for users. Now, in theory, we only need to configure in one place.
There’s already a bunch of articles in the Google Cloud Publication describing how to get started with the Antigravity suite.
Here I’m going to briefly share my experience in configuring MCP servers and skills.
This is literally day 1 of me experimenting with these things in Agy, so there’s a lot I need to discover. And I expect we’ll continue to see a lot of bug fixes across the Agy suite too. But I wanted to share this with you ASAP, as I’ve been getting a lot of folks asking me questions about this. I hope it’s useful for you.
As before, we have a .gemini folder in our home directory. But now it looks like this:
.gemini
├── antigravity/
├── config/
│ ├── plugins/ # Global / shared plugins
│ ├── projects/ # Approved project folders
│ └── mcp_config.json # Global / shared MCP config
├── antigravity-cli/
│ ├── mcp/ # Dynamically generated from shared MCP config
│ └── settings.json # CLI settings
├── antigravity-ide/
│ ├── mcp/ # Dynamically generated from shared MCP config
│ ├── plugins/ # Symlinked from global config
│ └── installation_id
├── skills/ # Global / shared skills
│ └── {skill_name}/
│ └── SKILL.md
└── GEMINI.md
Some things to note about this:
~/.gemini/antigravity/skills/<skill-folder>/
.gemini folder. We can confirm this by running the /skills command from inside Agy CLI:
~/.gemini/antigravity-cli/skills, whereas “shared” skills — meaning those available to **all workspaces running any Antigravity tools ** — should go in ~/.gemini/skills.mcp_config.json file and place it in the new ~/.gemini/config folder. (This is where “shared” extensions — now called “plugins” go also.)Here’s what my new ~/.gemini/config/mcp_config.json currently looks like:
{
"mcpServers": {
"lyria": {
"command": "mcp-lyria-go",
"env": {
"MCP_SERVER_REQUEST_TIMEOUT": "55000",
"GENMEDIA_BUCKET": "my-bucket-name",
"PROJECT_ID": "my-google-project",
"LOCATION": "europe-west1"
},
"disabled": true
},
"avtool": {
"command": "mcp-avtool-go",
"env": {
"MCP_SERVER_REQUEST_TIMEOUT": "55000",
"PROJECT_ID": "my-google-project",
"GENMEDIA_BUCKET": "my-bucket-name",
"LOCATION": "europe-west1"
}
},
"chirp3-hd": {
"command": "mcp-chirp3-go",
"env": {
"MCP_SERVER_REQUEST_TIMEOUT": "55000",
"GENMEDIA_BUCKET": "my-bucket-name",
"PROJECT_ID": "my-google-project"
}
},
"gcloud": {
"command": "npx",
"args": [
"-y",
"@google-cloud/gcloud-mcp"
]
},
"notebooklm-mcp": {
"command": "uvx",
"args": [
"--from",
"notebooklm-mcp-cli",
"--with",
"fakeredis<2.20.0",
"notebooklm-mcp"
],
"disabled": true
},
"StitchMCP": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://stitch.googleapis.com/mcp",
"--header",
"X-Goog-Api-Key: <my-stitch-api-key>"
],
"env": {}
},
"FirestoreMCP": {
"serverUrl": "https://firestore.googleapis.com/mcp",
"authProviderType": "google_credentials",
"oauth": {
"scopes": [
"https://www.googleapis.com/auth/cloud-platform"
]
},
"headers": {
"X-Goog-User-Project": "dazbo-portfolio"
},
"disabled": true
},
"remote-github": {
"serverUrl": "https://api.githubcopilot.com/mcp/",
"headers": {
"Authorization": "Bearer <my-github-api-key>",
"Content-Type": "application/json"
}
},
"google-developer-knowledge": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://developerknowledge.googleapis.com/mcp",
"--header",
"X-Goog-Api-Key: <my-google-developer-knowledge-api-key>"
],
"env": {}
},
"google-cloud-resource-manager": {
"serverUrl": "https://cloudresourcemanager.googleapis.com/mcp",
"authProviderType": "google_credentials"
}
}
}
A few notes about this MCP JSON file…
serverUrl, not the older httpUrlLet’s see if this works…

Let’s test one of these.
I say to Agy CLI: “List my public GitHub repos. Present them in a markdown table, along with their summaries, primary technologies, and number of stars. List in desc stars order.”
And Agy CL comes back with:

So that worked fine.
But what about in Agy IDE? Let’s just ask the agent, “What MCP servers do we have?”

So you can see that Agy IDE has picked up the same MCP configuration. Hurrah!
I did discover one issue… In my old Gemini CLI MCP server configurations, I was able to pass in environment variables, e.g. for API keys. In Agy CLI / IDE, this isn’t working for me and I’ve had to hard-code these keys into my mcp_config.json file. I hope this will get fixed soon.
I started by downloading and installing a number of skills in the usual way; i.e. via npx, like this:
# Skills related
npx skills add https://github.com/vercel-labs/skills -y -g --skill find-skills
# My skills - e.g. project-documentation
npx skills add https://github.com/derailed-dash/dazbo-agent-skills -y -g
# Agentic dev, Gemini, ADK
npx skills add https://github.com/google/skills -y -g
npx skills add https://github.com/GoogleCloudPlatform/vertex-ai-creative-studio/experiments/mcp-genmedia/skills -y -g
# Research, content, documentation, and marketing
npx skills add https://github.com/shubhamsaboo/awesome-llm-apps/awesome-agent-skills -y -g --skill deep-research
npx skills add https://github.com/shubhamsaboo/awesome-llm-apps/awesome-agent-skills -y -g --skill fact-checker
npx skills add https://github.com/shubhamsaboo/awesome-llm-apps/awesome-agent-skills -y -g --skill strategy-advisor
npx skills add https://github.com/shubhamsaboo/awesome-llm-apps/awesome-agent-skills -y -g --skill technical-writer
npx skills add https://github.com/wshobson/agents -y -g --skill architecture-decision-records
npx skills add https://github.com/coreyhaines31/marketingskills -y -g --skill seo-audit
# UI and frontend
npx skills add https://github.com/vercel-labs/agent-skills -y -g --skill vercel-react-best-practices
npx skills add https://github.com/remotion-dev/skills -y -g --skill remotion-best-practices
These commands install skills to your ~/.agents/skills folder. But I want them to be in ~/.gemini/skills. As mentioned previously, this location is picked up by all Agy products. So, go ahead and move the skills you’ve installed. Or you can run a script like this to do it for you:
for d in ~/.agents/skills/*/; \
do [ -d "$d" ] && rm -rf "$HOME/.gemini/skills/$(basename "$d")"; \
done && mv ~/.agents/skills/* ~/.gemini/skills/
By the way, I’ve actually created a skill that does this for me. I.e. it helps with installing any skill, and then relocating it to the appropriate location for Antigravity. It’s called deploy-skills-in-antigravity and you can check it out here. After installing it, I can say to Agy:
“please install the skill seo-audit from https://github.com/coreyhaines31/marketingskills”
Agy then activates my skill and uses the correct commands to install and move the newly installed SEO skill.

(If you like this skill, please add a star to my repo!)
When we start Agy CLI, these skills are all available. Let’s check in Agy IDE too. I ask Agy IDE: “What skills do we have installed?”

So we can see the skills are detected in Agy IDE, as required.
If I ask the Agy agent to “Perform a core documentation review,” we can see that it uses the required skills automatically:

(If you’re interested in this documentation review skill, then you can obtain it here. Please add a star to the repo, if you find it useful!)
So, where does that leave us?
Migrating from the old Gemini CLI to the new Antigravity ecosystem does require a bit of mental rewiring when it comes to folder structures. But having a single, unified source of truth for your MCP servers and skills is a leap forward. No more duplicating configurations or wondering why a custom skill works in your terminal but fails in your IDE. (Hopefully!)
Is it perfect? Not yet. We still have some day-one paper cuts — like that frustrating environment variable bug in the global MCP configuration that forces us to hardcode API keys for now. But given the sheer startup speed of the new Go-based CLI and the potential of the shared agent harness, the foundation is incredibly solid.
Hopefully, this guide saves you some time scratching your head over directory discrepancies. Let me know if you run into any other teething issues!
Now, if you’ll excuse me, I think it’s time to put on a Hawaiian shirt, mix up a piña colada, and go enjoy a bit of sunshine.
Until next time, happy building!
aiMost of us have seen a coding agent fail to complete a task we know it can do. We just don't...
googlecloudWhen building Generative AI applications, developers often encounter a massive bottleneck: sequential...
discussI’ve been thinking about sharing some electronic circuit posts on Dev.to — small circuits, DIY...
agentsWhat nobody tells you about exporting your multi-agent prototype to a local workspace. Every...
agenticarchitectAutonomous agents are genuinely good at answering messy business questions. Give one an LLM and a set...
aiPR volume went up, ticket quality didn't, and the gap got filled with LLMs on both sides of the review: bots reviewing, bots replying, bots occasionally arguing with bots about priorities that only existed in a teammate's head. Our CEO named the actual problem, and it's bigger than code review.
Workflows from the Neura Market marketplace related to this Stable Diffusion resource