Engineer Builds Allowlist-Only MCP Server to Give AI Safe VPS Access Without a Shell

·3 views

A developer has built a production-minded MCP (Model Context Protocol) server that allows an AI assistant to inspect a VPS without exposing an unrestricted shell. The server sits between the AI client and the VPS, permitting only a fixed set of read-oriented operations such as checking disk usage, container logs, SSL expiry, and service status. Each permitted operation maps to a pre-approved script on the server, and all requests must pass structured JSON validation before any SSH connection is made. Access is further restricted through a dedicated Linux user with limited sudo privileges and root-owned allowlisted scripts acting as a command gateway. The project, built in Python using the official MCP SDK, AsyncSSH, and Pydantic, prioritises operational visibility for AI tools while keeping arbitrary command execution entirely out of reach.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Replaces sklearn Pipeline With Rust, Cuts Docker Image Size 400x

A developer named Murat Genç published a technical article on DEV Community on August 1 detailing how he replaced a scikit-learn machine learning pipeline with one written in pure Rust. The rewrite resulted in a dramatic reduction in Docker image size, shrinking it from 900MB down to just 8MB — a roughly 400x decrease. The article falls under the topics of machine learning, Rust, and data science. The post is estimated to be a 7-minute read and received 3 reactions from the community.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Survey: Marketing Teams Using More AI Are Hiring More Staff, Not Fewer

A survey of over 1,000 marketing budget decision-makers by Exploding Topics found that 60% of marketing teams grew their headcount over the past year, while only 8.9% reduced it. Among teams most heavily investing in AI, the hiring trend was even stronger, with over 82% reporting headcount increases. The findings, published in a report updated April 1, 2026, push back against the widespread assumption that AI-driven automation leads to workforce reductions in marketing. Researchers attribute the growth to rising demand for staff who can manage expanded tool stacks, oversee AI-generated content quality, and handle strategic decisions that automation cannot replace. The survey notes these are correlations based on self-reported data, not proof that AI directly caused every hiring decision.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How LLMs and Playwright Can Build Self-Healing TypeScript Web Scrapers

Web scraping pipelines frequently break when front-end engineers make minor changes such as renaming CSS classes, restructuring the DOM, or updating React components, causing automation scripts to fail entirely. Traditional scrapers rely on rigid CSS selectors or XPath expressions that treat websites as static systems, making them highly vulnerable to the dynamic nature of modern web development. A new architectural approach combines Large Language Models with visual grounding, Model Context Protocol tool standardization, and WebGPU Compute Shaders to build TypeScript agents capable of recovering from such failures automatically. When a selector breaks, the agent captures a visual snapshot of the page, analyzes its spatial layout through multimodal processing, and dynamically recalculates its execution path. This self-healing design mirrors how intelligent API Gateways in microservices architectures adapt to upstream changes without disrupting downstream consumers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer shares 8 Astro optimizations that achieved perfect 100 mobile PageSpeed score

A web developer building a client site on the Astro framework achieved a perfect score of 100 across all four Google PageSpeed Insights categories on mobile by applying a series of incremental performance improvements. Key fixes included self-hosting web fonts with font-display swap, inlining stylesheets to eliminate render-blocking network requests, and using proper srcset attributes with responsive images. The developer also added fetchpriority hints for the Largest Contentful Paint image and configured long-term cache headers for hashed build assets. Astro's Islands Architecture, which ships minimal JavaScript by default, provided a strong performance baseline before any manual optimizations were applied. The findings highlight that achieving top Lighthouse scores typically requires combining multiple small fixes rather than relying on any single change.

0 comments Read more at DEV Community