April 20, 2026
Opus 4.7, Claude Design, and an MCP flaw affecting 200k servers
Get the next issue
TL;DR
Anthropic shipped Opus 4.7 — the first Opus with a 1M token context window and reduced latency. Claude Design launched in beta, letting teams generate production-ready UI components from screenshots or descriptions. A critical prompt injection flaw in a widely-used MCP base library was disclosed, affecting an estimated 200k server deployments. Plus: skills.sh V2 API, and the community's first Claude Code speed run.
API & Model Updates
Anthropic released Claude Opus 4.7 with a 1 million token context window (up from 200k in 4.5) and a ~30% reduction in time-to-first-token vs 4.5. Pricing: $15/M input, $75/M output — same as 4.5. Extended context is priced at $30/M beyond 200k tokens.
Claude Design is Anthropic's new product for generating production-ready React/Tailwind components from design screenshots, wireframes, or text descriptions. Outputs are editable in a live preview. Currently in closed beta — invite-only via the console.
MCP Ecosystem
A critical vulnerability was disclosed in mcp-base-server (versions <2.3.1), the most widely used Node.js base library for building MCP servers. A malicious tool description string could inject arbitrary instructions into the model's context. Patch: npm update mcp-base-server.
Claude Code & CLI
A developer livestreamed building a full SaaS product (auth, billing, CRUD, deploy) in 47 minutes using Claude Code with a CLAUDE.md template and a set of pre-built skills. The recording is the most-watched Claude Code video to date.
Developer Tools
skills.sh V2 API — programmatic install, search, publish
skills.sh launched a V2 REST API enabling programmatic skill search, install, publish, and version management. CI/CD pipelines can now auto-install skills as part of a build step. Rate limit: 1000 req/day free tier.
Action Items
- →npm update mcp-base-server immediately if you ship MCP servers
- →Apply for Claude Design beta via Anthropic console if you build UI
- →Try Opus 4.7 for long-document analysis tasks — 1M context is live
All Resources
If you found this useful, subscribe to get the next issue.
Get the next issue