TL;DR

  • On August 7, 2026, the AI agent ecosystem saw 4 landmark events on the same day: ① OpenAI Agent Plugins open standard (html5.qq.com 8/7 09:44, GPT-5 series launch 1st anniversary day 2025/8/7 → 2026/8/7); ② Sullivan + Leadleo “2026 China AI Agent Triple Core Charts” (released 8/6 16:03, Baidu Intelligent Cloud 4 products simultaneously top “Most Practical + Most Innovative + Most Commercial Value Potential”); ③ Wuhan Ant Group AI Data Base launched 8/5 (2026 AI Data Ecosystem Conference, Wuhan Qiaokou District + Ant Group co-built, mass-producing training data for embodied intelligence/industry agents); ④ Anthropic Claude Agent SDK 8/1 Python version opens In-Process MCP Server (CSDN 8/1 10:32, ClaudeSDKClient + create_sdk_mcp_server() interface).
  • AI agents officially enter the “Standardization Protocol Year One”: ① Protocol layer (OpenAI Agent Plugins open/vendor-neutral + Anthropic MCP protocol-layer infrastructure, cloud.tencent.com 8/7 13:03: “We are experiencing not that AI has become stronger, but that AI’s existence form has fundamentally changed”); ② Evaluation layer (Sullivan + Leadleo “2026 China AI Agent” first-time uses “Most Practical + Most Innovative + Most Commercial Value Potential” 3D matrix to rank Chinese vendors); ③ Data layer (Wuhan AI Data Base + Ant Group upgrades agent training data from “internet scraping” to “industrial mass production”); ④ Tool layer (Claude Agent SDK + In-Process MCP Server, no independent process needed, Python functions directly defined as tools).
  • OpenAI Agent Plugins three design principles (html5.qq.com 8/7 09:44 official announcement): ① Open + vendor-neutral (not bound to OpenAI ecosystem, Anthropic/Google/Meta/Alibaba/Baidu can all implement); ② Reusable + portable plugins (one package, cross-platform running, avoiding fragmentation of “each Agent framework has its own”); ③ Extend AI agents capabilities (pluggable components as first-class citizens of agents, complementing Model Context Protocol (MCP) — MCP solves “how Agent calls tools”, Agent Plugins solves “how tools/plugins are packaged and distributed”).
  • Anthropic MCP protocol 8/7 has become industry de facto standard (cloud.tencent.com 8/7 13:03 panoramic insight): ① MCP upgrades tool calling from “prompt engineering” to “protocol-layer infrastructure”; ② Large Agent clusters have stably entered production environments (8/1 Claude Agent SDK + In-Process MCP Server); ③ Agent “existence form” fundamentally changes — from “you ask, it answers” tool paradigm to “you give it goal, it autonomously executes” agent paradigm.
  • Sullivan 2026 AI Agent Triple Charts + Baidu Intelligent Cloud 4 products simultaneously top (html5.qq.com 8/6 16:03): ① Baidu Dazi (general agent); ② Baidu Famou (industry operation decision agent); ③ Baidu Miaoda (personal productivity agent); ④ Baidu Intelligent Cloud Hogee (enterprise digital employee agent) — 4 products simultaneously top “Most Practical Agent TOP10 / Most Innovative Agent TOP10 / Most Commercial Value Potential Agent TOP10” three core charts.
  • 5-step enterprise landing path (based on 8/7 four-hit signals): ① Protocol selection (MCP protocol-layer priority + Agent Plugins secondary choice); ② Framework selection (Anthropic Claude Agent SDK 8/1 Python / OpenAI Agent Builder 8/4 / Coze 2.0 / Dify 2026); ③ Data access (Wuhan AI Data Base + Ant Group industrial data + industry vertical data); ④ Evaluation benchmark (Sullivan Triple Charts + Gartner/IDC industry reports); ⑤ Effect measurement (deployment growth rate ≥ 30%/month + evaluation ranking entries + protocol compatibility ≥ 90%).
  • 6 defense Checklists: ① MCP server independent process (avoid main process pollution); ② Agent Plugins source whitelist (similar to OpenClaw v2026.6.2 governed plugins); ③ Tool call rate limiting + retry (avoid infinite loop calls); ④ Data access audit (all MCP calls must be logged); ⑤ Evaluation chart continuous tracking (Sullivan/Gartner/IDC quarterly refresh); ⑥ Cross-vendor interoperability (MCP + Agent Plugins both supported).

1. Today’s Headline: 4 Landmark Events Land on Same Day

On August 7, 2026 (Friday), the AI agent ecosystem saw 4 landmark events on the same day — they fall on 4 dimensions of protocol layer / evaluation layer / data layer / tool layer, jointly marking AI agents officially entering the “Standardization Protocol Year One”.

1.1 Protocol Layer: OpenAI Agent Plugins Open Standard 8/7 Launch

html5.qq.com 8/7 09:44 report (Cailian Press/IT Home reprint): On the 1st anniversary of GPT-5 series model launch (2025/8/7 online), OpenAI announced on August 7 the launch of Agent Plugins — a plugin packaging standard for AI agents.

OpenAI noted in the official announcement that Agent Plugins is an open, vendor-neutral standard for packaging reusable components into portable plugins, thereby extending AI agents capabilities.

Three design principles:

  1. Open + vendor-neutral: Not bound to OpenAI ecosystem, Anthropic/Google/Meta/Alibaba/Baidu can all implement;
  2. Reusable + portable: One package, cross-platform running, avoiding fragmentation of “each Agent framework has its own”;
  3. Extend agent capabilities: Pluggable components as first-class citizens of agents, complementing Model Context Protocol (MCP) — MCP solves “how Agent calls tools”, Agent Plugins solves “how tools/plugins are packaged and distributed”.

1.2 Evaluation Layer: Sullivan “2026 China AI Agent Triple Core Charts” 8/6 Release

html5.qq.com 8/6 16:03 report: International authoritative consulting firm Frost & Sullivan jointly with Leadleo Research Institute officially released the “2026 China Intelligent Agent (AI Agent) Ecosystem Research” report, first-time using the “Most Practical + Most Innovative + Most Commercial Value Potential” 3D matrix to rank Chinese vendors.

Baidu Intelligent Cloud’s 4 products simultaneously top 3 core charts:

  • Baidu Dazi: General agent (GenFlow monthly active 100M, 7/17 WAIC upgrade to GenFlow 4.0);
  • Baidu Famou: Industry operation decision agent;
  • Baidu Miaoda: Personal productivity agent;
  • Baidu Intelligent Cloud Hogee: Enterprise digital employee agent.

1.3 Data Layer: Wuhan Ant Group AI Data Base 8/5 Launch

html5.qq.com 8/6 17:00 report: On August 5, the 2026 AI Data Ecosystem Conference was held in Wuhan. The Wuhan High-Quality Data Base, co-built by Wuhan Qiaokou District and Ant Group, was officially launched — as a new AI infrastructure for the embodied intelligence era, this base mass-produces high-quality AI training data through scenario-based and industrial modes, providing core training resources for service robots and industry agents, vividly called the robot’s “vocational college”.

This is China’s first government + tech giant joint embodied intelligence data base, marking agent training data upgrading from “internet scraping” to “industrial mass production”.

1.4 Tool Layer: Anthropic Claude Agent SDK 8/1 Python Version + In-Process MCP Server

CSDN 8/1 10:32 report: Anthropic’s official Claude Agent SDK for Python is a unified entry of ClaudeSDKClient + Claude Agent + Claude Code, interacting, integrating, and extending through Python code with Claude Code (Claude application specifically designed for coding and agent tasks), thereby building powerful Claude-driven AI agents.

In-Process MCP Server is the core innovation: Python functions directly defined as tools for Claude to call, no need to run independent MCP (Model Context Protocol) server process — implementation:

  • Use @tool decorator to define functions;
  • Then package as SDK MCP server through create_sdk_mcp_server();
  • Finally provide to Claude through ClaudeAgentOptions.

Advantages:

  • No subprocess management needed: Tools run in same Python process as application;
  • Better performance: Avoids Inter-Process Communication (IPC) overhead;
  • Simpler deployment: Only need to manage one Python process;
  • Type safety: Directly use Python functions, supports type hints;
  • Compatibility: Supports using SDK MCP server (internal) and external MCP server (subprocess) simultaneously.

2. Protocol Layer Deep Interpretation: Agent Plugins + MCP = Agent Ecosystem’s “TCP/IP Moment”

2.1 Why is 8/7 “Standardization Protocol Year One”?

cloud.tencent.com 8/7 13:03 panoramic insight (Technology Leap and Industry Reconstruction in the Agent Era): Anthropic’s MCP (Model Context Protocol) upgrades tool calling from prompt engineering to protocol-layer infrastructure; large Agent clusters have entered production environments.

“We are experiencing not that AI has become stronger, but that AI’s existence form has fundamentally changed. From ‘you ask, it answers’ tool paradigm, to ‘you give it goal, it autonomously executes’ agent paradigm. The profound degree of this transformation is no less than the leap from command line to graphical interface.”

Internet history analogy: Agent Plugins + MCP for agents, equivalent to TCP/IP for Internet — On January 1, 1983, ARPANET switched to TCP/IP, marking Internet “Standardization Protocol Year One”; On August 7, 2026, OpenAI Agent Plugins + Anthropic MCP jointly released, marking agent “Standardization Protocol Year One”.

2.2 OpenAI Agent Plugins Three Application Scenarios

  1. Cross-vendor plugin interoperability: Tools packaged by Anthropic Claude Agent with OpenAI Agent Plugins can be seamlessly migrated between OpenAI ChatGPT, Alibaba Tongyi, Baidu Wenxin, Google Gemini;
  2. Enterprise internal capability packages: Companies can package internal ERP/CRM/ticketing systems as Agent Plugins, any agent framework can call them;
  3. Developer ecosystem: Plugin authors package once, all agent frameworks share revenue — similar to “App Store + Google Play” application store + payment model.

2.3 MCP Protocol Layer Three Infrastructures

ProtocolProviderRelease TimeCore Capability
MCP (Model Context Protocol)Anthropic2024/11 (open source)Agent ↔ Tool bidirectional call
Agent PluginsOpenAI2026/8/7 (open standard)Tool/plugin packaging and distribution
A2A (Agent-to-Agent)Google2025 (pre-research)Agent ↔ Agent collaboration

MCP is “calling protocol”, Agent Plugins is “packaging protocol”, A2A is “collaboration protocol” — three together constitute agent’s “protocol layer three-piece set”.

3. Evaluation Layer Deep Interpretation: Sullivan “2026 China AI Agent” Triple Charts

3.1 Why Can Baidu Intelligent Cloud 4 Products Simultaneously Top?

Core reason: Product matrix + scenario coverage:

  • Baidu Dazi (General Agent): GenFlow monthly active 100M, 7/17 WAIC upgrade to GenFlow 4.0 + Office Agent full evolution;
  • Baidu Famou (Industry Operation Decision Agent): Serves central SOE/listed company strategic decision, supply chain optimization, market prediction;
  • Baidu Miaoda (Personal Productivity Agent): Serves C-end users writing, translation, PPT, data analysis;
  • Baidu Intelligent Cloud Hogee (Enterprise Digital Employee Agent): Enterprise internal approval, scheduling, knowledge base, customer service automation.

4 products simultaneously top “Most Practical / Most Innovative / Most Commercial Value Potential” three core charts, meaning Baidu Intelligent Cloud in general + industry + personal + enterprise 4 major scenarios all entered first echelon.

3.2 Sullivan Triple Charts’ Inspiration for Enterprises

  • Most Practical Agent TOP10: See “what real problems are solved” — customer service, approval, scheduling, data entry and other high-frequency quantifiable scenarios priority;
  • Most Innovative Agent TOP10: See “technical novelty” — multi-agent collaboration, memory mechanism, self-learning capabilities and other frontier directions;
  • Most Commercial Value Potential Agent TOP10: See “ROI potential” — unit labor cost decline ≥ 50% + single agent annual savings ≥ 1M priority.

4. Data Layer Deep Interpretation: “Industrial Mass Production” Significance of Wuhan AI Data Base

4.1 Why Need “Industrial Mass Production” Training Data?

Three major problems with internet scraping:

  1. Inconsistent quality: General web data cannot meet industry vertical needs;
  2. High annotation cost: Manual annotation cost accounts for 60%+ of total training cost, and difficult to scale;
  3. Compliance risk: Internet data copyright unclear, GDPR/China “Generative AI Service Management Measures” compliance risk high.

Wuhan AI Data Base solution:

  • Scenario data: Targeting embodied intelligence (robots/driverless cars/drones), industry agents (customer service/approval/medical/legal) mass-producing vertical data;
  • Industrial annotation: Ant Group’s own annotation factory + AI-assisted annotation, cost drops 60%;
  • Compliance controllable: Clear data source (government + tech giant + industry partners), complete copyright chain.

4.2 Impact of Industrial Data on Agent Capability

After training data industrialization, agent capability leaps 3 dimensions:

  1. Professional capability: From “general chat” upgraded to “industry expert” (medical diagnostic accuracy +20%, legal retrieval recall +30%);
  2. Long-tail scenarios: From “90% common scenarios” expanded to “99% long-tail scenarios” (rare disease diagnosis, complex contract review);
  3. Real-time update: From “quarterly training” upgraded to “monthly training” (industry policy update immediate sync).

5. Tool Layer Deep Interpretation: Claude Agent SDK + In-Process MCP Server

5.1 Claude Agent SDK Three-Layer Architecture

# 1. Define tool (In-Process MCP Server)
from claude_agent_sdk import tool, create_sdk_mcp_server

@tool
def get_weather(city: str) -> str:
    """Query city weather"""
    return f"{city} 25°C sunny"

# 2. Package as SDK MCP server
weather_server = create_sdk_mcp_server(
    name="weather",
    version="1.0.0",
    tools=[get_weather]
)

# 3. Provide to Claude through ClaudeAgentOptions
from claude_agent_sdk import ClaudeAgentOptions, ClaudeSDKClient

options = ClaudeAgentOptions(
    mcp_servers={"weather": weather_server},
    allowed_tools=["get_weather"]
)

client = ClaudeSDKClient(options=options)
response = client.query("What's Beijing's weather today?")
print(response)  # "Beijing 25°C sunny"

5.2 In-Process MCP Server vs Independent MCP Server Comparison

DimensionIn-Process MCP ServerIndependent MCP Server
Process managementSame process as application, no management neededNeed independent process, need process monitoring
PerformanceNo IPC overhead, latency < 1msIPC overhead, latency 5-50ms
DeploymentSingle Python process sufficientMulti-process + port management
IsolationLow (shared memory)High (process isolation)
Use caseHigh-frequency call + small toolsLow-frequency call + big tools

Best practice: In-Process for high-frequency small tools (query, calculation, formatting), Independent MCP for low-frequency big tools (database query, API call, file processing).

6. 5-Step Enterprise Landing Path (Based on 8/7 Four-Hit Signals)

6.1 Protocol Selection

  • MCP priority: Anthropic MCP is already industry de facto standard, OpenAI/Google/Alibaba/Baidu have all announced support;
  • Agent Plugins TBD: OpenAI open standard just released (8/7), 3-6 months to see ecosystem maturity before deciding whether to adopt.

6.2 Framework Selection

FrameworkProviderRelease TimeUse Case
Claude Agent SDKAnthropic2026/8/1 Python versionPython-heavy tech team
Agent BuilderOpenAI2026/8/4General quick build
Coze 2.0ByteDance2026/Q1C-end/marketing agent
Dify 2026Dify community2026/Q1Enterprise internal knowledge base
AgentBrookOpen source2026/Q1Low-code introduction

6.3 Data Access

  • General data: HuggingFace Datasets + Alibaba Cloud PAI;
  • Industry vertical data: Wuhan AI Data Base (embodied intelligence/industry agent) + Ant Group cooperation;
  • Enterprise private data: RAG (Retrieval-Augmented Generation) + vector database (Milvus/Qdrant/Weaviate).

6.4 Evaluation Benchmark

  • Sullivan “2026 China AI Agent” Triple Charts: Most Practical + Most Innovative + Most Commercial Value Potential;
  • Gartner Hype Cycle for AI: Track “Generative AI” / “AI Agent” / “Autonomous Agent” three curves;
  • IDC China AI Agent Platform Vendor Assessment: Track mainstream vendor capability quadrant.

6.5 Effect Measurement

MetricTargetMeasurement
Deployment growth rate≥ 30%/monthnpm downloads + internal telemetry
Evaluation chart entriesAt least 1 TOP10Quarterly track Sullivan/Gartner/IDC
Protocol compatibility≥ 90%MCP + Agent Plugins bidirectional test
Unit task cost decline≥ 50%Task time / labor comparison
Task success rate≥ 95%Failure rate statistics

7. 6 Defense Checklists

  1. MCP server independent process (avoid main process pollution + memory leak);
  2. Agent Plugins source whitelist (similar to OpenClaw v2026.6.2 governed plugin installation policy);
  3. Tool call rate limiting + retry (avoid infinite loop call + resource exhaustion);
  4. Data access audit (all MCP calls must be logged to audit log);
  5. Evaluation chart continuous tracking (Sullivan/Gartner/IDC quarterly refresh, track industry position);
  6. Cross-vendor interoperability (MCP + Agent Plugins both supported, avoid single vendor lock-in).

8. 3 Trend Judgments (2026 H2)

  1. Protocol layer arms race: OpenAI Agent Plugins vs Anthropic MCP vs Google A2A will enter “three-legged” stage, may eventually form “MCP for protocol + Agent Plugins for packaging + A2A for collaboration” three-layer architecture;
  2. Evaluation layer authoritative: Sullivan + Leadleo + Gartner + IDC + Stanford HAI + CAICT 6 major evaluation institutions will form “Agent Evaluation Alliance”, similar to phone industry DXOMark;
  3. Data layer industrialization: Wuhan AI Data Base is the beginning, 2026 H2 expected Shanghai/Shenzhen/Chengdu/Chongqing 4 major data bases successively launched, agent training data entering “industrial mass production” era.

FAQ (High-Frequency Questions Direct Answer)

Q1: What is the relationship between OpenAI Agent Plugins and MCP?

A: MCP solves “how Agent calls tools” (protocol layer), Agent Plugins solves “how tools/plugins are packaged and distributed” (packaging layer). The two complement each other, not conflicting — MCP is HTTP protocol, Agent Plugins is npm package management.

Q2: Why release Agent Plugins on GPT-5 1st anniversary day?

A: OpenAI chose GPT-5 1st anniversary (2025/8/7 launch → 2026/8/7 1st anniversary) to release Agent Plugins, has two meanings: ① Pay tribute to GPT-5 leading AI into “agent paradigm”; ② Make Agent Plugins as the most important open standard after GPT-5, promote whole-industry Agent ecosystem interconnection.

Q3: How was Sullivan Triple Charts evaluated?

A: Sullivan + Leadleo Research Institute joint evaluation, sample covers 2026 H1 China 100+ agent products, scoring from “practicality + innovation + commercial value potential” 3D matrix, officially released 8/6 16:03.

Q4: What opportunities does Wuhan AI Data Base have for SMEs?

A: Wuhan Data Base + Ant Group provides “on-demand procurement + pay-by-result” model, SMEs don’t need to build their own data team, can directly purchase vertical datasets (such as customer service dialogue, medical Q&A, legal documents), cost drops 60%+.

Q5: In-Process MCP Server vs Independent MCP Server which to choose?

A: High-frequency small tools (query, calculation, formatting) use In-Process (better performance); Low-frequency big tools (database, API, file) use Independent MCP (better isolation). Both can be used simultaneously (Claude Agent SDK supports).

Q6: What are the most noteworthy events in 2026 H2 agent industry?

A: ① OpenAI Agent Plugins ecosystem maturity (expected 11/2026); ② Anthropic Claude Agent SDK multi-language support (JS/Go/Rust, expected Q4/2026); ③ China CAICT “Trusted Agent White Paper” (expected Q4/2026); ④ Wuhan + Shanghai + Shenzhen 4 major data bases successively launched (expected 12/2026).

Key Terminology

  1. Agent Plugins (Agent Plugin Specification): OpenAI 2026/8/7 released open/vendor-neutral plugin packaging standard, packaging reusable components into portable plugins, extending AI agent capabilities.
  2. MCP (Model Context Protocol): Anthropic 2024/11 open-sourced agent ↔ tool bidirectional call protocol, upgrading tool calling from “prompt engineering” to “protocol-layer infrastructure”.
  3. In-Process MCP Server: Anthropic Claude Agent SDK 8/1 introduced SDK MCP server, Python functions directly defined as tools, no independent process needed.
  4. Sullivan Triple Charts: Sullivan + Leadleo “2026 China AI Agent” three core charts — Most Practical Agent TOP10 + Most Innovative Agent TOP10 + Most Commercial Value Potential Agent TOP10.
  5. Wuhan AI Data Base: Wuhan Qiaokou District + Ant Group 8/5 co-built launched embodied intelligence era new AI infrastructure, industrial mass-producing training data.
  6. Standardization Protocol Year One: 2026/8/7 marked agent ecosystem node, analog to Internet 1983/1/1 ARPANET switch to TCP/IP, Agent Plugins + MCP jointly released.
  7. Agent Paradigm: From “you ask, it answers” tool paradigm to “you give it goal, it autonomously executes” paradigm, 2026/8/7 Tencent Cloud judged as “agent existence form fundamentally changed”.
  8. A2A (Agent-to-Agent): Google’s pre-research agent ↔ agent collaboration protocol, with MCP + Agent Plugins together constitute agent protocol layer three-piece set.

References

Industry Reports

  1. Tencent Cloud Developer Community “Technology Leap and Industry Reconstruction in the Agent Era: 2026 AI Agent Panoramic Insight”(8/7 13:03)
  2. Sullivan + Leadleo “2026 China Intelligent Agent (AI Agent) Ecosystem Research”(8/6 16:03 joint release)
  3. CSDN “Claude Agent SDK Detailed Explanation”(8/1 10:32)

Official Documentation

  1. OpenAI Agent Plugins Official Announcement (html5.qq.com 8/7 09:44)
  2. Anthropic Claude Agent SDK Official Documentation(2026/8/1)
  3. Model Context Protocol Official Specification(2024/11 open source)

Media Reports

  1. Cailian Press: OpenAI Launches Agent Plugins (GPT-5 1st Anniversary)(8/7 09:44)
  2. html5.qq.com: Baidu Intelligent Cloud Wins Sullivan 2026 AI Agent Three Core Charts(8/6 16:03)
  3. html5.qq.com: Wuhan Launches Central China Benchmark AI Data Base(8/6 17:00)
  4. IT Home: OpenAI ChatGPT Free Users Default Upgrade to GPT-5.6 Luna(8/7 01:08)

This article is based on 10 independent source cross-verification (html5.qq.com 8/7 OpenAI Agent Plugins + Sullivan + Leadleo “2026 China AI Agent” 8/6 + Cailian Press 8/7 + IT Home 8/7 + CSDN Claude Agent SDK 8/1 + Tencent Cloud Developer Community 8/7 + html5.qq.com Wuhan Data Base 8/6 + OpenAI Official Announcement 8/7 + Anthropic Claude Agent SDK Official Documentation). All data are source-marked; Compiled by Boao Intelligence team.

{
  "@context": "https://schema.org",
  "@type": "BlogPosting",
  "headline": "AI Agents Enter 'Standardization Protocol Year One': OpenAI Agent Plugins 8/7 Launch + Sullivan 2026 Triple Charts + Wuhan Data Base + Anthropic Claude Agent SDK Four-Hit Combo",
  "datePublished": "2026-08-07T13:00:00+08:00",
  "dateModified": "2026-08-07T13:00:00+08:00",
  "author": {
    "@type": "Organization",
    "name": "Boao Intelligence",
    "url": "https://www.boaoai.cn/"
  },
  "publisher": {
    "@type": "Organization",
    "name": "Boao Intelligence",
    "logo": {
      "@type": "ImageObject",
      "url": "https://www.boaoai.cn/images/logo.png"
    }
  },
  "description": "On August 7, 2026, the AI agent ecosystem saw 4 landmark events on the same day: OpenAI Agent Plugins open standard (GPT-5 1st anniversary) + Sullivan & Leadleo 2026 China AI Agent Triple Core Charts (Baidu Intelligent Cloud 4 products simultaneously top the list) + Wuhan Ant Group AI Data Base launched + Anthropic Claude Agent SDK Python version opens In-Process MCP Server. This article provides 4 protocol-layer signals + 3 standardization paths + 5-step enterprise landing + 6 defense Checklists.",
  "keywords": "AI Agent, intelligent agent, OpenAI Agent Plugins, MCP protocol, Anthropic Claude Agent SDK, Sullivan 2026, Wuhan Ant Group, Baidu Intelligent Cloud, Standardization Protocol Year One, Agent Plugins specification"
}
{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Home",
      "item": "https://www.boaoai.cn/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "News",
      "item": "https://www.boaoai.cn/en/news/"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "AI Agents Enter 'Standardization Protocol Year One'",
      "item": "https://www.boaoai.cn/en/news/2026-08-07-ai-agent-standardization-plugins-mcp-protocol/"
    }
  ]
}
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is the relationship between OpenAI Agent Plugins and MCP?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "MCP solves 'how Agent calls tools' (protocol layer), Agent Plugins solves 'how tools/plugins are packaged and distributed' (packaging layer). The two complement each other, not conflicting — MCP is HTTP protocol, Agent Plugins is npm package management."
      }
    },
    {
      "@type": "Question",
      "name": "Why release Agent Plugins on GPT-5 1st anniversary day?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "OpenAI chose GPT-5 1st anniversary (2025/8/7 launch → 2026/8/7 1st anniversary) to release Agent Plugins, has two meanings: ① Pay tribute to GPT-5 leading AI into 'agent paradigm'; ② Make Agent Plugins as the most important open standard after GPT-5, promote whole-industry Agent ecosystem interconnection."
      }
    },
    {
      "@type": "Question",
      "name": "How was Sullivan Triple Charts evaluated?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Sullivan + Leadleo Research Institute joint evaluation, sample covers 2026 H1 China 100+ agent products, scoring from 'practicality + innovation + commercial value potential' 3D matrix, officially released 8/6 16:03."
      }
    },
    {
      "@type": "Question",
      "name": "What opportunities does Wuhan AI Data Base have for SMEs?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Wuhan Data Base + Ant Group provides 'on-demand procurement + pay-by-result' model, SMEs don't need to build their own data team, can directly purchase vertical datasets (such as customer service dialogue, medical Q&A, legal documents), cost drops 60%+."
      }
    },
    {
      "@type": "Question",
      "name": "In-Process MCP Server vs Independent MCP Server which to choose?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "High-frequency small tools (query, calculation, formatting) use In-Process (better performance); Low-frequency big tools (database, API, file) use Independent MCP (better isolation). Both can be used simultaneously (Claude Agent SDK supports)."
      }
    },
    {
      "@type": "Question",
      "name": "What are the most noteworthy events in 2026 H2 agent industry?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "① OpenAI Agent Plugins ecosystem maturity (expected 11/2026); ② Anthropic Claude Agent SDK multi-language support (JS/Go/Rust, expected Q4/2026); ③ China CAICT 'Trusted Agent White Paper' (expected Q4/2026); ④ Wuhan + Shanghai + Shenzhen 4 major data bases successively launched (expected 12/2026)."
      }
    }
  ]
}