95% recall, 80% answers: what LongMemEval retrieval scores don't tell you
Memory layers for AI agents love to quote LongMemEval. You will see 94%, 95%, 96%. Those numbers are real, but they usually measure retrieval: was o…
AI & ML
⚑ Report a ProblemLatest AI & ML news from Tech News
All topics
AI
News
Tech
agents
ai
api
architecture
automation
backend
beginners
career
claude
database
devchallenge
devops
javascript
llm
machinelearning
mcp
opensource
productivity
programming
python
react
security
showdev
testing
tutorial
typescript
webdev
Замена Lens, которую целиком написала модель: что получилось и где мы проигрываем
Sextant — бесплатный open-source клиент Kubernetes на Rust и Tauri 2: всё, что вы делали в Lens, но без аккаунта, подписки и телеметрии. Его код цел…
Your MCP Hub Exposes Every Tool to Every Agent. That's the Bug.
We stood up an internal MCP hub last quarter. It aggregated four services behind one endpoint, listed every tool in a single catalog, and let any co…
Как превратить Cursor в долговременную память проекта: LLM Wiki, QMD и Cursor CLI
Большой контекст в Cursor спасает ровно до тех пор, пока проект живёт дольше одного чата: требования, SQL-гипотезы и причины отброшенных решений быс…
Pin baseUrl in the Gateway, Not in the Agent's Prompt
The other day an agent I was testing kept hitting production. Not on purpose, not because it was being clever. It was reading the OpenAPI spec we ha…
When the Spec Changes, the MCP Tool List Should Change With It
The first MCP server we stood up, someone on the team wrote out the tool list by hand. A JSON file with 22 entries, each describing one endpoint we…
The bug is in the arrangement: finding and cutting your AI agent's data-leak paths
Give an agent three ordinary tools: one that reads a GitHub issue, one that reads a local folder, one that can post a GitHub comment. Each behaves e…
Can AI Agents Safely Write to the General Ledger? 13 Write Operations Counted, 0 Silent in a Locked Period
Can AI Agents Safely Write to the General Ledger? 13 Write Operations Counted, 0 Silent in a Locked Period Yes — if the write surface is bounded twi…
Сутки из жизни G3A1SF: как я учу агента без backprop — днём он живёт, ночью учится
Что происходит с агентом, который не решает одну задачу, а живёт — долго, в мире, который всё время меняется? Я собрал NOWA-Cognitive, когнитивную а…
Can AI Agents Access Your HR and Payroll Data Safely? 329 Tools Counted, 54 Write/Delete
Can AI Agents Access Your HR and Payroll Data Safely? 329 Tools Counted, 54 Write/Delete Yes — if every request carries the user's permissions, and…
Figma MCP + Claude Code: Give It Your Real Layout, Variables, and Components
Paste a screenshot of a Figma frame into Claude Code and you get something that looks close: hard-coded hex colors, guessed spacing, a brand-new but…
Your MCP Server Returns the Source. Does the Answer Still Cite It?
A benchmark for what happens to citations after a retrieval server hands its results to an AI agent it doesn't control. Part 1 of 6 in the series Ci…
Designing an MCP server for a personal data app: 6 lessons from 10 wardrobe servers
The Model Context Protocol (MCP) lets an AI assistant such as ChatGPT or Claude call tools on your server. Several closet apps already have an MCP s…
Triglav: orchestrating Claude Code and Codex agents for long-running work and layered review
AI coding agents can write code. Longer tasks still need coordination and quality control. Someone has to keep track of the objective, settle the op…
My AI agent read a web page on Monday and got tricked on Tuesday. So I built a ledger.
I've been working on a side project called ADTL (Agent Data Trust Labels) for a while now. It just hit v0.4, and the new part is something I didn't…
Vaultic: local-first encrypted memory for MCP-compatible AI agents
AI agents are useful, but starting from zero in every conversation makes continuity difficult. Vaultic is a local-first memory layer for MCP-compati…
I built an MCP server that lets AI agents book real hotel rooms (no OTA commission)
Booking.com takes 15-25% commission and owns the customer relationship. I run a small network of independent establishments in Luxembourg, and I wan…
An agent timeout is not a cancelled action: building Mothership
An agent asks to save a report. The request needs your approval. You step away, the client times out, and the server restarts. When you return, shou…
I Finally Understood What MCP Actually Does in Claude Code
When I first started using Claude Code, I treated MCP as another feature I probably didn't need yet. I could already ask Claude to read my code, mod…
Building a Google Merchant Center MCP server: what the Merchant API doesn't tell you
I run GSC Wizard , an SEO tool with an MCP server that already exposed Search Console and GA4 to Claude and ChatGPT. This month I added Google Merch…
Записать — не значит вспомнить: месяц внешней памяти для ИИ-агента
Однажды я потерял чат. Не то чтобы катастрофа — обычная рабочая сессия по проверке студенческих домашек. Экспорт я не делал, в лог ничего не писал.…
Deploying MCP Servers on Kubernetes
Most MCP servers start life as a local, stdio-based process — the client spawns the server as a child process, messages flow over stdin and stdout.…
AI Growth Engine Competitor Website Watch Agent: Architecture, Tool Schemas & Integration Guide
Why conversational agent prompts fail in production—and the exact typed tool schemas, MCP servers, and dry-run safeguards required for Grok Bot Arch…
11 MCP Servers That Prove MCP Isn't Dead
Disclosure: I'm the founder of PDFHaul, which is #5 on this list. Originally published on the PDFHaul blog , where the full source list with links l…
Langflow pipes commands into bash
CVE-2026-105740 lets any authenticated Langflow user run arbitrary commands on the host, and the cause is about as plain as RCE gets: the command fi…
Сколько инструментов должно быть у MCP-сервера? Я измерила от 5 до 100 в Claude Code, Codex и Gemini
Почти все советы по проектированию MCP-серверов сводятся к одному: держите инструментов поменьше. Я проанализировала 91 источник. 38 из них советуют…
MCP-сервер с авторизацией: что мы сделали, чтобы это было безопасно
Привет! У нас новость: ВкусВилл приготовил MCP сервер с авторизацией! Адрес тот же, mcp.vkusvill.ru/mcp , а для доступа к полному функционалу пройди…
REA Hit 248K Stars. The Real Story Is Scarier.
REA Hit 248K Stars. The Real Story Is Scarier. A week ago, morluto/rea had fewer than 10,000 GitHub stars. Today it has over 248,000 — making it one…
Give Claude or Cursor read-only access to public Instagram data with one MCP URL
If you ask an AI assistant about an Instagram account, it usually guesses. So it can't help with a competitor or a creator you want to research. We…
One MCP config for Claude Code, OpenCode and OMP
Three files for one server You use Claude Code for one task and OpenCode for another, and both should see the same Figma and Linear MCP servers. Eac…