[KongchangAI]

#MCP

516 related articles

How Many Tokens Do MCP Tool Definitions Cost? Testing Four Servers Against the CLI
·4 min

How Many Tokens Do MCP Tool Definitions Cost? Testing Four Servers Against the CLI

Tested four MCP servers' tool definition token costs: GitHub's default config uses 11,207 tokens, Grafana 20,493. JSON Schema accounts for 83%, descriptions just 12%. With tool search on, resident cost drops to a few hundred tokens — the CLI isn't cheaper.

Read more →
Claude Code in Practice: AI Programming from Installation to Full-Stack Project Development
·5 min

Claude Code in Practice: AI Programming from Installation to Full-Stack Project Development

Based on Mosh Hamadani's Claude Code course, this guide covers installation and setup, prompt techniques, Plan Mode, context window management, MCP extensions, and engineering methodology for bug fixing and refactoring—helping you build production-grade full-stack apps from scratch.

Read more →
Two Essential Updates for Building Chrome Extensions with Antigravity
·2 min

Two Essential Updates for Building Chrome Extensions with Antigravity

Two key updates for building Chrome extensions with Antigravity: install the Modern Web Guidance best-practices skill package and the Chrome DevTools for Agents MCP server, enabling AI Agents to install and test extensions in a real browser for a complete code-to-verification loop.

Read more →
The Underrated AI Agent Tool: Apify MCP Lets Agents Scrape the Entire Web
·3 min

The Underrated AI Agent Tool: Apify MCP Lets Agents Scrape the Entire Web

AI Agents struggle to scrape Google Maps, Instagram, and LinkedIn. Apify MCP Server connects to Claude with one line of config, tapping 80,000+ ready-made scrapers — five scenarios tested for just $2.71, with a full setup guide and advanced Skill combinations.

Read more →
When Should You Not Build an MCP Server? AI Engineer Interview Question Explained
·2 min

When Should You Not Build an MCP Server? AI Engineer Interview Question Explained

AI engineer interview question: When should you NOT build an MCP server? Same-process single function calls don't need MCP; too many tools consume context tokens and degrade model performance; fixed workflows should be coded directly.

Read more →
Tool Descriptions Are Protocol: 7 Key Questions in MCP Tool Design
·2 min

Tool Descriptions Are Protocol: 7 Key Questions in MCP Tool Design

A tool description is part of the model's calling protocol, not traditional documentation. This article breaks down the field responsibilities of name, description, schema, and annotations in MCP tool design, and offers 7 pre-publish check questions to help developers eliminate silent failures.

Read more →
Finally Understanding What MCP Actually Does in Claude Code
·2 min

Finally Understanding What MCP Actually Does in Claude Code

A developer's take on what MCP actually does in Claude Code: it doesn't make Claude smarter, it lets it reach GitHub, databases, browsers, and other external systems. Plus the common myth that more MCP servers is better.

Read more →
Plunger: See Exactly What Your AI Agent Is Sending to APIs
·3 min

Plunger: See Exactly What Your AI Agent Is Sending to APIs

Plunger is an open-source HTTP client for AI agents with GUI, CLI, and MCP entry points sharing requests and history, offering variable validation against placeholder leaks, credential masking, and tamper-evident logs for observable, auditable API calls.

Read more →
AI Agent Auto-Fills Timesheets: Webhook + MCP Power Law Firm Billing Drafts
·2 min

AI Agent Auto-Fills Timesheets: Webhook + MCP Power Law Firm Billing Drafts

The AI-MCP Agent platform shows how Webhook triggers and MCP tool connections build an AI Agent that auto-drafts billable consultant time entries — no chat interface, launched by enterprise end-of-day systems, generating reviewable time entries from calendars and messages.

Read more →
RESTk + MCP in Practice: Let AI Agents Operate Your API Client Directly
·3 min

RESTk + MCP in Practice: Let AI Agents Operate Your API Client Directly

RESTk uses a local MCP Server to let Claude Code, Cursor, and other AI assistants operate the API client directly, supporting automated audits, doc generation, and Agent-readiness scoring. Fully local, credential-redacted, approval-gated—demonstrating a new paradigm for AI Agent and dev tool integration.

Read more →
What Makes Up an Agentic AI System? Five Core Components
·5 min

What Makes Up an Agentic AI System? Five Core Components

Read more →
MCP vs Function Calling: A Decision Framework for Agent Tool Selection
·2 min

MCP vs Function Calling: A Decision Framework for Agent Tool Selection

MCP vs LLM function calling: compare core differences, MCP's hidden operational costs, and a practical decision framework for agent tool selection.

Read more →
MCP: The New Standard for AI Tool Integration — Are You Still Not On Board?
·3 min

MCP: The New Standard for AI Tool Integration — Are You Still Not On Board?

MCP (Model Context Protocol) is emerging as the unified standard for AI tool integration. This article covers its architecture principles, comparison with custom integrations, common first-week mistakes, and the vendor landscape.

Read more →
Laravel Meets MCP Servers: A Practical Guide from Install to Test
·2 min

Laravel Meets MCP Servers: A Practical Guide from Install to Test

Hands-on Laravel MCP package tutorial: setup and install, creating MCP servers and tools via Artisan, local vs remote registration, testing with the built-in inspector, and best practices for descriptions and safety annotations.

Read more →
New Ideas for Agent Memory: The More You Use It, the Better You Remember
·2 min

New Ideas for Agent Memory: The More You Use It, the Better You Remember

Agent Commons launches Agent notes: topic-based storage, tag retrieval, and read-count reinforcement. Frequently recalled memories rank higher—designed as "a notebook, not a vector database," available via HTTP, MCP, and A2A.

Read more →

Building a Privacy-First Medical Imagi…

·6 min

Building a Privacy-First Medical Imaging AI Agent with Claude and MCP

Build a privacy-first medical imaging AI agent with Claude and MCP: Aegis local de-identification and six narrow tools keep PHI and pixels local while the model only orchestrates.

Read more →
A2A vs MCP vs ACP: The Protocol War Is Over, But Teams Are Still Paying for It
·4 min

A2A vs MCP vs ACP: The Protocol War Is Over, But Teams Are Still Paying for It

What's the difference between A2A, MCP, and ACP? ACP has been merged into A2A and is now history. This article analyzes the real cost of the protocol war, the two-layer MCP and A2A stack, what enterprise-ready really means, and the trade-offs you must accept when choosing a protocol.

Read more →
Chatway MCP: Let ChatGPT and Claude Take Over Customer Support Tickets Directly
·2 min

Chatway MCP: Let ChatGPT and Claude Take Over Customer Support Tickets Directly

Chatway MCP lets AI assistants like ChatGPT and Claude manage support tickets, reply to messages, search contacts, and access knowledge bases via the MCP protocol—no code required.

Read more →