Welcome to the daily AI news brief for vibe coders. It is Friday, July 3, 2026, and the last 24 hours favored the tools you build with. Google made Firebase Studio natively agentic, Anthropic hardened the Claude Agent SDK and shipped Claude in Chrome to general availability, and OpenAI pushed GPT-5.6 Sol onto Cerebras. If you ship on Shopify, Claude, Firebase, or GPT, today is a tooling day.
TL;DR
- Firebase Studio adds Agent modes, native MCP support, and Gemini CLI; Genkit gains early Python and expanded Go.
- Claude Agent SDK adds a dedicated skills option and a tools option; MCP tunnels reach research preview.
- Claude in Chrome is now generally available; Claude Code adds a /dataviz skill and draft PR handoff.
- Claude Sonnet 5 is the default model for every Free and Pro user worldwide.
- OpenAI runs GPT-5.6 Sol on Cerebras at up to 750 tokens per second for select customers.
- Build of the day: connect one private MCP server to your agent using the new MCP tunnels preview.
Firebase Studio goes natively agentic
What shipped. Google expanded Firebase Studio, its cloud workspace for full-stack AI apps, with new Agent modes, foundational Model Context Protocol support, and Gemini CLI integration. Genkit added early Python support and broadened Go, and templates for Flutter, Angular, React, Next.js, and web gained Gemini capabilities (Google Developers Blog).
Why it matters for vibe coders. If your stack runs on Firebase and Genkit, orchestration now sits closer to your deploy target, and native MCP reuses the tool servers you run in Claude Code.
What to do today. Open a workspace, switch on an Agent mode, and point it at a Genkit flow to test the MCP wiring.
Claude Agent SDK and MCP get control and security upgrades
What shipped. The Claude Agent SDK deprecated the old “Skill” entry in allowed_tools for a dedicated skills option, and added a tools option to control the base tool set. The MCP dependency floor was raised to patch a disclosed vulnerability, and MCP tunnels reached research preview so agents can reach private-network servers (Releasebot).
Why it matters for vibe coders. Granular skills and tools controls make agent behavior more predictable, and private tunnels expose internal tools without publishing them.
What to do today. Update the SDK, migrate allowed_tools Skill entries to the skills option, and confirm your MCP dependency meets the new minimum.
Claude in Chrome hits general availability
What shipped. Anthropic released Claude in Chrome as generally available and expanded Claude Code with background notifications, draft pull-request handoff, improved failover, and better session handling. Claude Code also added a /dataviz skill for chart and dashboard design guidance (Releasebot).
Why it matters for vibe coders. Browser-native Claude acts on rendered pages, which helps when testing storefronts or reading output raw fetches miss. Draft PR handoff keeps a human reviewing agent code.
What to do today. Run the /dataviz skill on a real dataset, then ship a small agent change as a draft PR, not a direct commit.
Claude Sonnet 5 is now the default for everyone
What shipped. Claude Sonnet 5 became the default model for every Free and Pro user worldwide, positioned near Opus 4.8 on agentic work at introductory pricing of $2 per million input tokens through August 31. Reported agentic coding scores reached 63.2 percent, up from 58.1 percent (Build Fast with AI).
Why it matters for vibe coders. A Sonnet-class model this close to Opus shifts the cost-to-capability math for everyday coding, and the pricing window makes it cheap to standardize now.
What to do today. Run one real ticket through Sonnet 5, compare it to your current default, and check token counts before switching.
OpenAI runs GPT-5.6 Sol on Cerebras at 750 tokens/sec
What shipped. OpenAI began serving GPT-5.6 Sol on Cerebras at up to 750 tokens per second for select customers as capacity expands. Sol is the flagship of the GPT-5.6 family, alongside lower-cost Terra and faster Luna, previewed to 20 partners with US government coordination (VentureBeat).
Why it matters for vibe coders. High throughput changes what feels interactive; at 750 tokens per second, agent loops and long generations move closer to real time.
What to do today. Watch for general availability, and design latency-sensitive flows so you can swap the endpoint later.
Also worth noting
- Google shipped two Gemini image models: Gemini 3.1 Flash Image at $0.50 per million input tokens and $3.00 output, and Gemini 3 Pro Image at $2.00 input and $12.00 output (Build Fast with AI).
- Reflection AI’s $6.3 billion compute lease at SpaceX’s Colossus 2 in Memphis went live July 1 at $150 million per month for Nvidia GB300 access through 2029 (Build Fast with AI).
- California Governor Gavin Newsom signed a deal giving state agencies, cities, and counties access to Claude at a 50 percent discount (Build Fast with AI).
- Claude Fable 5 returned to users worldwide on July 1 after a 19-day export-control suspension was lifted (Build Fast with AI).
Build of the day
Wire one internal tool into an agent using MCP tunnels. Pick a small private service, such as an inventory lookup or order-status endpoint. Stand up a minimal MCP server in front of it, then connect your agent through the MCP tunnels research preview so it can call the tool without exposing the service publicly. Test one round trip and log the call. That is a proof of concept in under two hours, and a template for every internal capability you expose next.
FAQ
What is the fastest way to try Firebase Studio’s new agentic features?
Open a Firebase Studio workspace, enable an Agent mode, and connect a Genkit flow. It now includes foundational Model Context Protocol support and Gemini CLI integration, so you can reuse MCP tool servers you already run. Start with one small flow before you add it to a production project (Google Developers Blog).
Do I need to change my Claude Agent SDK code today?
If you set tools through the allowed_tools Skill entry, migrate to the new skills option, which the SDK now favors for granular control. Also raise your MCP dependency to the new minimum to pick up the disclosed security fix. Both changes are small, and making them now avoids a forced migration later (Releasebot).
What does Claude in Chrome reaching general availability mean for me?
General availability means Claude in Chrome is broadly usable, not limited to a preview cohort. Browser-native Claude can act on rendered pages, which helps when testing storefronts or reading output raw fetches miss. Claude Code also added draft pull-request handoff, so agent changes arrive as reviewable drafts rather than direct commits (Releasebot).
Is Claude Sonnet 5 worth switching to for coding?
Sonnet 5 is now the default for Free and Pro users and sits near Opus 4.8 on agentic tasks, with reported agentic coding scores of 63.2 percent. Introductory pricing is $2 per million input tokens through August 31. Run one real ticket through it against your current model and check token counts before you standardize (Build Fast with AI).
How fast is GPT-5.6 Sol on Cerebras, and can I use it now?
OpenAI reports GPT-5.6 Sol on Cerebras at up to 750 tokens per second, initially for select customers as capacity expands. The GPT-5.6 family was previewed to 20 partners with US government coordination, with wider availability planned. Design latency-sensitive flows so you can swap the endpoint in later without rewriting orchestration (VentureBeat).
Which Gemini image model should I use for a Shopify store?
Route drafts to Gemini 3.1 Flash Image at $0.50 per million input tokens and $3.00 output, then escalate approved concepts to Gemini 3 Pro Image at $2.00 input and $12.00 output for final assets. The two-tier split keeps costs low during iteration while reserving the higher-fidelity model for store-facing visuals (Build Fast with AI).
Sources
- Google Developers Blog — Firebase Studio AI capabilities
- Releasebot — Anthropic updates
- Releasebot — Claude Code updates
- Build Fast with AI — AI News July 1, 2026
- VentureBeat — OpenAI GPT-5.6 Sol, Terra, Luna
About the author
Robert McCullock runs Design Delight Studio, building AI-assisted commerce on Shopify with multi-agent architectures, Claude Code, and Firebase. See selected work in his professional portfolio.