Article URL: https://lucumr.pocoo.org/2026/7/4/better-models-worse-tools/ Comments URL: https://news.ycombinator.com/item?id=48788599 Points: 16 # Comments: 5
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Apple shipped an official Safari MCP server with Safari Technology Preview 247 in early July 2026, marking the first time a major browser vendor has natively integrated Model Context Protocol support for AI-driven debugging. The server is built on safaridriver and exposes 17 tools covering navigation, DOM inspection, element interaction, network capture, console logging, and screenshots. It runs entirely on the user's machine with no data sent to Apple, and each AI session launches in an isolated window with no access to personal browser data such as cookies, logins, or autofill. Before this release, all MCP browser automation tools relied on Chromium, forcing Mac developers who prefer Safari to run a second browser solely for AI agent tasks. The server is currently available only in Safari Technology Preview and not yet in the stable Safari release.
Infrastructure as Code (IaC) configurations written in Terraform can carry serious security vulnerabilities, just like application code, according to a developer experiment published on DEV Community. The author deliberately wrote an insecure AWS Terraform setup featuring a public S3 bucket, open security groups, an unencrypted database with a hardcoded password, and a wildcard IAM admin policy. Running Checkov, an open-source SAST tool maintained by Prisma Cloud with over 1,000 built-in policies, against just 70 lines of code surfaced 35 failed security checks in seconds without requiring any AWS credentials. The author then remediated all 35 issues and integrated the Checkov scan into a GitHub Actions CI pipeline to catch misconfigurations automatically before deployment. Similar real-world misconfigurations have been linked to major data breaches, including incidents involving Capital One and exposed US voter records.
A frontend developer shares their firsthand exploration of agentic software development, a growing approach where AI is given broader objectives rather than single-task prompts. Unlike traditional AI interactions that require a developer to initiate each step, AI agents operate in a continuous loop — planning, executing, and evaluating progress until a goal is met. The developer notes that tools like this could automate repetitive tasks such as setting up project structures, freeing engineers to focus on product thinking and user experience. Despite the shift, the author argues that developers remain essential for understanding requirements and ensuring the right solutions are delivered. The key takeaway is that AI is evolving from answering questions to completing entire software workflows, though human judgment and problem-solving remain irreplaceable.
PaperQuire has released version 0.3.0, introducing a built-in Model Context Protocol (MCP) server that allows AI agents like Claude, ChatGPT, and Copilot to convert Markdown directly into formatted PDF documents. MCP is an open protocol enabling AI applications to call external tools through a standard JSON-RPC interface, eliminating the need for manual copy-paste and formatting steps. The MCP server exposes four tools — render, list_templates, show_template, and batch_render — and works with the same Chromium-based rendering engine used in PaperQuire's desktop app. Users can configure PaperQuire as an MCP server in Claude Desktop, VS Code, or any compatible client by adding a short JSON snippet to their config file. The app is available for macOS, Windows, and Linux, with a free tier offering three renders per day and a Pro plan for unlimited use.
Discussion (0)
Log in to join the discussion and vote.
Log in