惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

GbyAI
GbyAI
Simon Willison's Weblog
Simon Willison's Weblog
Microsoft Security Blog
Microsoft Security Blog
Y
Y Combinator Blog
The GitHub Blog
The GitHub Blog
Engineering at Meta
Engineering at Meta
F
Fortinet All Blogs
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
A
About on SuperTechFans
Last Week in AI
Last Week in AI
月光博客
月光博客
有赞技术团队
有赞技术团队
P
Proofpoint News Feed
MyScale Blog
MyScale Blog
Martin Fowler
Martin Fowler
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
C
Check Point Blog
U
Unit 42
The Register - Security
The Register - Security
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Hugging Face - Blog
Hugging Face - Blog
阮一峰的网络日志
阮一峰的网络日志
V
Visual Studio Blog
酷 壳 – CoolShell
酷 壳 – CoolShell
D
DataBreaches.Net
WordPress大学
WordPress大学
aimingoo的专栏
aimingoo的专栏
H
Hacker News: Front Page
Recent Announcements
Recent Announcements
C
CXSECURITY Database RSS Feed - CXSecurity.com
Latest news
Latest news
小众软件
小众软件
P
Palo Alto Networks Blog
PCI Perspectives
PCI Perspectives
Security Latest
Security Latest
S
Secure Thoughts
Scott Helme
Scott Helme
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
T
Threat Research - Cisco Blogs
P
Proofpoint News Feed
M
MIT News - Artificial intelligence
Application and Cybersecurity Blog
Application and Cybersecurity Blog
Google DeepMind News
Google DeepMind News
Recorded Future
Recorded Future
O
OpenAI News
S
Securelist
云风的 BLOG
云风的 BLOG
H
Help Net Security
T
Troy Hunt's Blog

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant Common SOC 2 Failures (Real World) Stop Vibe-Checking Your AI App: A Practical Guide to Evals How to Use SonarQube and SonarScanner Locally to Level Up Your Code Quality Your Next To-Do App Is Dead — I Replaced Mine with an OpenClaw AI Sign a Nostr event in 60 lines of Python using coincurve — no nostr-sdk, no nbxplorer, no rust toolchain ITGC Audit Explained Like You’re in Big 4 Patch Tuesday abril 2026: Microsoft parcha 163 vulnerabilidades y un zero-day en SharePoint Stop scraping everything: a better way to track competitor price changes Listing on MCPize + the Official MCP Registry while routing payments OUTSIDE the marketplace — how I kept 100% of my x402 revenue Building an AI-Powered Risk Intelligence System Using Serverless Architecture Why We Ripped Function Overloading Out of Our AI Toolchain Testing AI-Generated Code: How to Actually Know If It Works SaaS Churn Is Killing Your Business. Here Is What to Do About It (Without a Support Team) The Speed of AI Is No Longer Linear - And Self-Improving Models Are Why How to Implement RBAC for MCP Tools: A Practical Guide for Engineering Teams From Standard Quote to Persuasive Proposal: AI Automation for Arborists I built a CLI that scaffolds complete multi-tenant SaaS apps Axios CVE-2025–62718: The Silent SSRF Bug That Could Be Hiding in Your Node.js App Right Now The dashboard that ended our friendship Data Pipelines Explained Simply (and How to Build Them with Python) The Hidden Cost of AI Systems Nobody Talks About. undefined vs undeclared, and how typeof behaves Switching from file-based jobs to NATS/Kafka in Rust without changing code io_uring Adventures: Rust Servers That Love Syscalls Why Agentic AI is Killing the Traditional Database The POUR principles of web accessibility for developers and designers Quantum Neural Network 3D — A Deep Dive into Interactive WebGL Visualization How To Install Caveman In Codex On macOS And Windows Automation Pipeline Reliability: Why Your Workflow Breaks When Nobody Is Watching I Built an 'Open World' AI Coding Agent — It Works From ANY Folder From Freelancing to Product: A Tech Service Company's SaaS Transformation China's AI Giants: Adding Tencent Hunyuan & ByteDance Doubao to AI University (74 Providers) On the Vibe Coders and Their Lies clerk: Auto-Summarize Your Claude Code Sessions AI Weekly — 2026/04/10–04/17 | The Model Lockdown Is Here, but the Toolchain Is the Real Battleground AI 週報 — 2026/04/10–2026/04/17 模型封鎖潮來了,但工具鏈才是真戰場 Maybe this is how Open-Source apps are born... 🚀 Fine-Tune LLMs with LoRA and QLoRA: 2026 Guide tRPC v11 + Next.js App Router: End-to-End Type Safety Without the Boilerplate ShadCN UI in 2026: Why I Stopped Installing Component Libraries and Started Owning My Components SaaS Billing in React Server Components: Stripe + Supabase Without a Single `useEffect` Join our DEV Weekend Challenge — $1,000 in Prizes Across TEN winners! Submissions Due April 20 at 6:59 AM UTC. Implementing FSRS Spaced Repetition in Flutter + Supabase — Adding Memory Science to an AI Learning App "I Texted My Localhost From the Train — Claude Code Fixed the Bug Before I Got Home" I Built a Sales Prep AI and It Went Deeper Than Expected Design to Code #2: One JSON, Eleven Outputs Solving the 100M-Row Problem: A Summary Table Pattern for High-Volume Push Notification Logs Flutter Web With Wasm: What Actually Changes For Developers I Built 50 Royalty-Free Soundtracks for My Side Project in a Weekend Using AI Music Generation The Vibe Coding Security Checklist: 7 Things to Check Before You Ship Stop Letting Googlebot Guess Fix Your React App's SEO Right Desconstruindo o Streaming do LinkedIn: Como Criar um Engine de Extração de Vídeo de Alta Performance com HLS e FFmpeg (EDA Part-1) EDA (Exploratory Data Analysis) Explained With Real Life — Why Looking at Your Data Is the Most Important Step in Machine Learning Brand Relationship Management at Scale: Our 4-Touch Outreach System for 200+ Brands Why String.fromEnvironment() Might Return an Empty String in Dart JGuardrails 1.0.0 — Hardening Java LLM Apps Against Jailbreaks, Toxicity, and Prompt Injection Plan and Schedule a Full Week of Threads Content From One Claude Conversation Coding Cat Oran Ep3, Five Tables Changed Everything Updated: BFF Pattern I'm done watching freelancers get buried by 200 proposals. So I'm building the alternative. This is my first post BFS Algorithm in Java Step by Step Tutorial with Examples Tracking LLM Pricing Monthly: An Open Dataset for 22 AI Models How We Measure Content ROI on a Comparison Site: Revenue Attribution Without Perfect Data Introducing Nova AI Ops: The AI-Native Operating System for SRE Teams I built a free desktop video downloader for Windows — Grabbit How Talkie OCR Helps Vision-Impaired & Dyslexic Users Read the World Around Them VRCFaceTracking安装和iPhone面捕配置教程,有bug Even CrowdStrike Can't See Your Agents The Automation Gold Rush: What n8n Workflows and Claude Are Opening Up for Developers Right Now
Seu agente de IA está desperdiçando 13.000 tokens antes de dizer "oi"
Rudson Kiyos · 2026-04-29 · via DEV Community

E você provavelmente nem sabe disso.


Se você tem um agente com 50 tools MCP instaladas, aqui está o que acontece antes de qualquer mensagem do usuário ser processada:

{
  "name": "gmail_send_email",
  "description": "Sends an email message via the Gmail API to one or more 
    recipients. Use this tool when the user explicitly requests to send, 
    compose and send, or deliver an email message to someone.",
  "input_schema": {
    "type": "object",
    "required": ["to", "subject", "body"],
    "properties": {
      "to": {
        "type": "string",
        "description": "The recipient email address or comma-separated list"
      },
      "subject": {
        "type": "string", 
        "description": "The subject line of the email"
      },
      "body": {
        "type": "string",
        "description": "The body content of the email in plain text or HTML"
      }
    }
  }
}

Enter fullscreen mode Exit fullscreen mode

Isso é ~195 tokens. Por ferramenta. Antes de qualquer coisa.

50 tools × 195 tokens = 9.750 tokens de overhead puro.

E isso é só o catálogo. Ainda não chegou no contexto do usuário, na memória da conversa, nos documentos, em nada.


"Mas tem prompt caching, não?"

Sim. E reduz o custo financeiro para ~10% do valor original.

Mas caching não reduz o custo de atenção.

Esses tokens continuam ocupando a janela de contexto. O modelo ainda processa tudo na atenção a cada request. E se você usa retrieval dinâmico de tools — selecionando ferramentas diferentes por request — o cache quebra em cada seleção diferente.

A conta não some. Ela só fica mais barata.


O problema real que ninguém fala

O MCP JSON Schema foi projetado como contrato de execução de ferramenta. Não como contrato semântico de seleção.

Resultado: informação crítica para o LLM raciocinar está ausente ou enterrada em texto livre:

  • Sem contrato de erro — o LLM não sabe o que fazer quando auth_failed
  • Sem trigger explícito — tem que inferir "quando usar essa tool" de uma description de parágrafo
  • Sem taxonomia de retrieval — não tem como agrupar ou filtrar tools por domínio

Ou seja: verboso E semanticamente incompleto. O pior dos dois mundos.


TTC — TERSE Tool Catalog

Passei as últimas semanas resolvendo esse problema. O resultado é uma extensão do TERSE Format chamada TTC — TERSE Tool Catalog.

A mesma ferramenta acima em TTC:

TOOL gmail_send_email
  PURPOSE: send email via Gmail
  IN: to:string, subject:string, body:string, cc:string?
  OUT: message_id:string
  ERR: auth_failed | quota_exceeded | invalid_recipient
  WHEN: user wants to send or compose an email
  TAGS: gmail, email, communication

Enter fullscreen mode Exit fullscreen mode

~55 tokens. Redução de 73.6%.

E repara no que foi adicionado, não só no que foi removido:

Campo MCP JSON TTC
ERR — contrato de falha ❌ ausente ✅ explícito
WHEN — trigger de seleção ❌ enterrado ✅ explícito
TAGS — taxonomia de retrieval ❌ ausente ✅ explícito

Não é compressão. É realocação.

Esse é o ponto mais importante do spec, e vale deixar claro:

TTC não economiza tokens removendo conteúdo semântico. Ele elimina overhead sintático e documental do JSON Schema — que serve legibilidade humana, não raciocínio de LLM — e reinveste parte dessa economia em semântica explícita de seleção de ferramentas.

A conta real:

MCP JSON Schema:        ~195 tokens por tool
TTC sem campos novos:    ~35 tokens
TTC com todos os campos: ~65 tokens

Os 30 tokens de "reinvestimento" compram:
  ERR  → contrato de falha (ausente no MCP)
  WHEN → trigger semântico (ausente no MCP)  
  TAGS → taxonomia de retrieval (ausente no MCP)

Resultado: 195 → 65 tokens. -66.6%.
Mas os 65 tokens carregam mais sinal de raciocínio
do que os 195 originais.

Enter fullscreen mode Exit fullscreen mode

É ganho líquido de sinal de raciocínio, não ganho de informação no sentido clássico.


Benchmark real — 10 tools medidas

Medi com tokenizador BPE (cl100k_base) em 10 definições reais de tools MCP:

Tool JSON Schema TTC Redução
gmail_send_email 208 55 73.6%
calendar_create_event 262 78 70.2%
github_create_issue 269 84 68.8%
jira_create_ticket 254 77 69.7%
slack_send_message 206 69 66.5%
Total (10 tools) 1.948 650 66.6%

Projeção para catálogos maiores:

Catálogo JSON Schema TTC Economia
20 tools ~3.896 ~1.300 ~2.596 tokens
50 tools ~9.740 ~3.250 ~6.490 tokens
100 tools ~19.480 ~6.500 ~12.980 tokens

A economia absoluta cresce linearmente. Quanto maior o catálogo, maior o ROI.


Vocabulário normativo para WHEN

Um campo de linguagem natural sem padrão cria outro problema: dois autores de servidores MCP diferentes escrevem WHEN de formas incompatíveis, degradando a acurácia de seleção em catálogos grandes.

O TTC v1.0 resolve isso com vocabulário normativo:

WHEN: user [wants|requests|asks|needs|intends] to [ação] [objeto]

Exemplos conformantes:
  WHEN: user wants to send an email message
  WHEN: user requests to list files in Google Drive
  WHEN: user needs to create a calendar event

Não-conformante:
  WHEN: send email          ← falta verbo de intenção
  WHEN: user email          ← falta verbo de ação

Enter fullscreen mode Exit fullscreen mode

Simulação de acurácia (TF-IDF cosine similarity, 12 tools, 36 queries):

Condição Acurácia
MCP description livre 63.9%
TTC WHEN vocabulário controlado 72.2%
Delta +8.3 pp

Caveat: simulação TF-IDF, não benchmark real com LLM. Evidência direcional.


Onde funciona melhor

Catálogos grandes (20+ tools) — onde a economia absoluta justifica a migração

Modelos locais e menores — Qwen 7B, Llama 3, Mistral — sem cache, janelas estreitas

Pipelines multi-agente — o overhead se acumula a cada passagem de contexto

RAG de tools — TTC compacto é ideal para indexar em vetor DB e injetar subsets

❌ Catálogos pequenos com LLM grande e contexto amplo — ganho marginal

❌ Substituir JSON Schema em contratos de API — não é o propósito


Links


Se o seu agente tem 50 tools instaladas e você ainda não pensou no custo de atenção do catálogo, esse é um bom momento para repensar.


Tags: ai agents mcp llm tooling performance opensource