// edition · 2026-05-26

May 26, 2026

24 stories on AI dev tools, agents, and the coding stack — curated from the day's RSS haul by Agentic Dev's pipeline.

Top Signal · MCP & Integrations

Claude MCP Tunnels: Private MCP Access for Agents

Anthropic announced MCP tunnels for Claude Managed Agents on May 19, 2026, as a research preview requiring access requests. The feature allows Claude agents to connect to private Model Context Protocol servers without public endpoints, using Cloudflare's cloudflared for outbound-only tunnel conne...

Dev.to - Claude

Tool Updates

How I manage 40+ skills across Claude Code, Codex, and .agents folders

A developer built a Go CLI tool to manage selective loading of 40+ skills across AI coding agents including Claude Code and Codex, after global skill directories caused irrelevant tool suggestions across different project contexts. Prior fixes including manual file moves and shell aliases failed ...

CLI Agents Dev.to - Claude

Vibe Coding: My Daily Workflow with Claude Code

A developer published their workflow for using Claude Code in AI-assisted ("vibe") coding, describing a hands-on approach where they write project plans independently before consulting the model, use separate context files to reduce token usage, and validate AI-suggested bug diagnoses before allo...

CLI Agents Dev.to - Claude

How I Cut My Anthropic API Bill by 50% With a Local Python Tool

A developer built a local Python CLI tool called `ai-cost-optimizer` that reduced their Anthropic API costs by 50% using a semantic cache, a BM25-based prompt compressor, and a model router that directs simple queries to cheaper models like Haiku ($1/1M tokens) instead of Opus ($15/1M tokens).

Workflows & Tips Dev.to - Claude

Add Runtime Limits to Claude Agent Workflows

A Dev.to guide describes a TypeScript pattern for adding runtime limits to Claude-based AI agent workflows, using constraints such as maximum execution time (30 seconds), step count (15), and tool calls (10) to prevent runaway retries and unbounded execution.

Agent Engineering Dev.to - Claude

Beyond the Prompt: Why Your AI Agent Needs a Governance Runtime

A developer building a product called NEES Core Engine argues that production AI agents require a dedicated governance runtime layer to enforce business logic and safety boundaries, rather than relying solely on system prompts. The article identifies failure modes including policy bypass, memory ...

Agent Engineering Dev.to - AI

The git Commands You Forgot Exist (And Why AI Workflows Make Them Relevant Again)

A technical guide covers underused git commands including `git worktree`, `git bisect`, `git rerere`, `git log -S`, `git notes`, and `git range-diff`. The article notes that `git worktree`, which allows multiple simultaneous working directories from one repo, has practical use in AI coding agent ...

Workflows & Tips Dev.to - Claude

GitLab 19.0 trades its string section for a full DevSecOps orchestra

GitLab released version 19.0 on May 21, 2026, introducing a Secrets Manager in public beta for Premium and Ultimate users that scopes credentials to individual CI/CD jobs. The release also adds agentic merge request workflows, CI pipeline visibility, and supply chain visibility features.

Agent Engineering The New Stack

How We Analyzed The Top 2,354 ClawHub Skills for Security

Trent AI analyzed 2,354 packages on ClawHub using a five-step behavioral pipeline that evaluates AI agent skills for permission scope, credential handling, network exposure, input validation, and chained attack paths, alongside VirusTotal scans as a secondary data point.

Agent Engineering Dev.to - AI

Microsoft Copilot Cowork Exfiltrates Files

Security researchers at PromptArmor disclosed a vulnerability in Microsoft Copilot's Cowork feature that allows attackers to exfiltrate files, likely via prompt injection techniques targeting the AI assistant's access to user documents.

Agent Engineering Hacker News - Best

Ecosystem

Two Models Just Hit 90% on Agent Coding. One Cost Less Than a Penny.

A benchmark across 10 agent coding tasks found Qwen3 Coder 30B A3B and DeepSeek Chat both scored 90%, at costs of $0.0004 and $0.0018 per run respectively. Liquid's LFM 2 24B A2B scored 85% at $0.0002, while Qwen3 14B scored 0% due to mandatory thinking-mode timeouts.

Model Releases Dev.to - AI

My AI corrections used to be tribal knowledge. I shipped the fix into my own OSS.

A developer released Hypomnema, an open-source personal wiki tool that integrates with Claude Code via 14 lifecycle hooks, available via `npm install -g hypomnema` under the MIT license. It stores notes as plain markdown with git, requires no vector database or API keys, and auto-manages session ...

Open Source Tools Dev.to - AI

MCP server with live prediction market data — one config line [87849]

A developer published a Nexus MCP server that connects Claude Desktop to live Kalshi prediction market data via a single JSON config line, covering symbols including BTC, ETH, FED, and CPI with 60-second updates. A paid arbitrage endpoint is available at $0.01 per request on the Base blockchain, ...

MCP & Integrations Dev.to - Claude

Voicebox: The Open-Source AI Voice Studio That Just Hit 28K Stars

Voicebox, an open-source desktop voice synthesis application, has reached 28,500 GitHub stars. Released under the MIT license, it runs locally, supports 7 text-to-speech engines, voice cloning from short audio clips, 23 languages, and includes a Model Context Protocol server for AI agent integrat...

Open Source Tools Dev.to - AI

Add live Fed rate cut probability to Claude Desktop — 30 second setup [11750]

A developer published a tutorial for adding a third-party MCP server to Claude Desktop that retrieves Federal Reserve rate cut probability data from prediction markets such as Kalshi. The service, called Nexus, also supports crypto and macro symbols, with a paid tier at $9.99/month via RapidAPI.

MCP & Integrations Dev.to - Claude

Using AI to write better code more slowly

Developer Nolan Lawson wrote about his approach of using AI tools to produce higher-quality code at a slower pace, arguing that deliberate AI-assisted development trades speed for improved output quality.

Opinion & Analysis Hacker News - Best

What Is an Agentic Browser? How AI Browsers Work Proactively for Users

Agentic browsers are a category of AI-enhanced web browsers designed to autonomously execute multi-step tasks based on natural language instructions, rather than requiring manual user input. Google Chrome has introduced an "Auto Browse" feature and Opera launched Opera Neon as examples of this ap...

Opinion & Analysis Dev.to - AI

AiFinPay: Autonomous Payments for ruvnet/ruflo

AiFinPay released a Python library called `aifinpay-agent` that provides payment infrastructure for AI agents. The package is installable via pip and the source code is available on GitHub under the AiFinPay organization.

Open Source Tools Dev.to - AI

You Cannot Outrun a Wave

Rick Houlihan, using Claude as a coding assistant, migrated his family's laundromat business infrastructure to Oracle Cloud Free Tier, building a stack he claims could support a 75-location franchise at near-zero cost.

Opinion & Analysis Dev.to - Claude

Uber president says AI spending is getting ‘harder to justify’

Uber president Andrew Macdonald said the company cannot draw a clear connection between rising AI spending — including token consumption for Claude Code — and useful consumer features delivered, after Uber reportedly exhausted its 2026 AI budget within four months.

Industry & Funding The Verge - AI