Blog

Notes on AI agents,
MCP and staying in the loop.

Practical guides to the Model Context Protocol, coding agents like Claude Code and Cursor, and the human-in-the-loop patterns that let you step away from the terminal without losing control.

Tutorial

Claude Code Notifications: Get Pinged When It Finishes or Needs You

Two hook events cover it: Notification fires when Claude Code is waiting on you, Stop fires when it finishes a turn. Here is the exact config for desktop and phone alerts, and why your hook seems late.

10 min read

Read
Tutorial

How to Connect a Remote MCP Server to Any AI Client

One remote MCP server, nine clients, nine slightly different config formats. Here is the exact file, key and header syntax for each, plus the fixes for the errors you will actually hit.

11 min read

Read
Guide

Best MCP Servers in 2026: 10 Worth Connecting, With Install Commands

Ten MCP servers worth connecting to Claude Code, Cursor or Codex in 2026, each with a verified install command, the switch that keeps it read-only, and what it costs.

14 min read

Read
Tutorial

Cursor Notifications: Get Alerted When the Agent Finishes or Waits

Cursor can chime, show a desktop banner, push to an iPhone, or run a script when the agent stops. Here is every option, what each one catches, and how to get a push on any phone.

9 min read

Read
Comparison

Claude Code Remote Control: What It Does, Its Limits, and Alternatives

Remote Control lets you drive a Claude Code session on your own machine from the Claude app or a browser. How to set it up, what it can't do, and when another tool fits better.

9 min read

Read
Security

Human-in-the-Loop AI Agents: What to Gate and What If Nobody Answers

Pausing an agent is easy. Asking rarely, asking clearly and deciding what happens when nobody answers is the hard part. A policy guide with OWASP mappings, data and code.

11 min read

Read
Comparison

ntfy vs Pushover vs Slack: Which Push Channel Fits Your AI Agents?

Three good ways to get a ping from a script or a coding agent, compared on price, limits, platforms and replies. Plus what to do when the agent needs an answer, not just a ping.

9 min read

Read
Tutorial

ChatGPT MCP: How to Add a Custom MCP Server

ChatGPT moved custom MCP servers from developer mode to the Plugins page. Here are the current steps, who can use them, and why ChatGPT keeps asking you to confirm.

9 min read

Read
Tutorial

LangGraph Human-in-the-Loop: interrupt(), Command and Phone Approvals

Pause a LangGraph run with interrupt(), resume it with Command(resume=...), and send the approval to your phone from a worker that survives restarts. Tested code throughout.

8 min read

Read
Explainer

What Is MCP? The Model Context Protocol, Explained for 2026

MCP is the open standard that lets AI apps like Claude, ChatGPT and Cursor discover and call external tools. How it works, what the 2026-07-28 spec changed, who governs it, and how to use it safely.

11 min read

Read
by PostMCP AI