作者：来自 Elastic Justin Castilla
Agent Builder 现在作为技术预览版提供。使用 Elastic Cloud 试用即可开始，并在这里查看 Agent Builder 文档。
A2A 和 MCP：代码实战
这是文章 “在 Elasticsearch 中使用 A2A 协议和 MCP 创建一个 LLM agent 新闻室” 的配套内容，该文章解释了在同一个 agent 中同时实现 A2A 和 MCP 架构以真正获得两个框架独特优势的好处。我们也提供了一个代码仓库，方便你自己运行这个演示。
让我们来看看新闻室的 agent 如何同时使用 A2A 和 MCP 协作完成一篇新闻文章。你可以在这里找到配套的代码仓库，查看这些 agent 的实际运作。
步骤 1：选题分配
新闻主管（作为 client）分配一个选题：
`
1. {
2. "message_type": "task_request",
3. "sender": "news_chief",
4. "receiver": "reporter_agent",
5. "payload": {
6. "task_id": "story_renewable_energy_2024",
7. "assignment": {
8. "topic": "Renewable Energy Adoption in Europe",
9. "angle": "Policy changes driving solar and wind expansion",
10. "target_length": 1200,
11. "deadline": "2025-09-30T18:00:00Z"
12. }
13. }
14. }
`AI写代码
步骤 2：记者请求调研
Reporter agent 发现它需要背景信息，并通过 A2A 将任务委派给 Researcher agent：
`
1. {
2. "message_type": "task_request",
3. "sender": "reporter_agent",
4. "receiver": "researcher_agent",
5. "payload": {
6. "task_id": "research_eu_renewable_2024",
7. "parent_task_id": "story_renewable_energy_2024",
8. "capability": "fact_gathering",
9. "parameters": {
10. "queries": [
11. "EU renewable energy capacity 2024",
12. "Solar installations growth Europe",
13. "Wind energy policy changes 2024"
14. ],
15. "depth": "comprehensive"
16. }
17. }
18. }
`AI写代码
步骤 3：记者向 Archive agent 请求历史背景
Reporter agent 意识到历史背景会强化报道。它通过 A2A 将任务委派给由 Elastic 的 A2A agent 驱动的 Archive agent，在新闻室的 Elasticsearch 驱动文章档案中进行搜索：
`
1. {
2. "message_type": "task_request",
3. "sender": "reporter_agent",
4. "receiver": "archive_agent",
5. "payload": {
6. "task_id": "archive_search_renewable_2024",
7. "parent_task_id": "story_renewable_energy_2024",
8. "capability": "search_archive",
9. "parameters": {
10. "query": "European renewable energy policy changes and adoption trends over past 5 years",
11. "focus_areas": ["solar", "wind", "policy", "Germany", "France"],
12. "time_range": "2019-2024",
13. "result_count": 10
14. }
15. }
16. }
`AI写代码
步骤 4：Archive agent 使用带 MCP 的 Elastic A2A agent
Archive agent 使用 Elastic 的 A2A agent，而这个 A2A agent 又使用 MCP 来访问 Elasticsearch 工具。这个过程展示了混合架构：A2A 负责 agent 协作，而 MCP 提供工具访问：
`
1. # Archive Agent using Elastic A2A Agent
2. async def search_historical_articles(self, query_params):
3. # The Archive Agent sends a request to Elastic's A2A Agent
4. elastic_response = await self.a2a_client.send_request(
5. agent="elastic_agent",
6. capability="search_and_analyze",
7. parameters={
8. "natural_language_query": query_params["query"],
9. "index_pattern": "newsroom-articles-*",
10. "filters": {
11. "topics": query_params["focus_areas"],
12. "date_range": query_params["time_range"]
13. },
14. "analysis_type": "trend_analysis"
15. }
16. )
18. # Elastic's A2A Agent internally uses MCP tools:
19. # - platform.core.search (to find relevant articles)
20. # - platform.core.generate_esql (to analyze trends)
21. # - platform.core.index_explorer (to identify relevant indices)
23. return elastic_response
`AI写代码
Archive agent 从 Elastic 的 A2A agent 收到全面的历史数据，并将其返回给 Reporter：
`
1. {
2. "message_type": "task_response",
3. "sender": "archive_agent",
4. "receiver": "reporter_agent",
5. "payload": {
6. "task_id": "archive_search_renewable_2024",
7. "status": "completed",
8. "archive_data": {
9. "historical_articles": [
10. {
11. "title": "Germany's Energiewende: Five Years of Solar Growth",
12. "published": "2022-06-15",
13. "key_points": [
14. "Germany added 7 GW annually 2020-2022",
15. "Policy subsidies drove 60% of growth"
16. ],
17. "relevance_score": 0.94
18. },
19. {
20. "title": "France Balances Nuclear and Renewables",
21. "published": "2023-03-20",
22. "key_points": [
23. "France increased renewable target to 40% by 2030",
24. "Solar capacity doubled 2021-2023"
25. ],
26. "relevance_score": 0.89
27. }
28. ],
29. "trend_analysis": {
30. "coverage_frequency": "EU renewable stories increased 150% since 2019",
31. "emerging_themes": ["policy incentives", "grid modernization", "battery storage"],
32. "coverage_gaps": ["Small member states", "offshore wind permitting"]
33. },
34. "total_articles_found": 47,
35. "search_confidence": 0.91
36. }
37. }
38. }
`AI写代码
这一步展示了 Elastic 的 A2A agent 是如何融入新闻室工作流的。Archive agent（新闻室的专用 agent）与 Elastic 的 A2A agent（第三方专家）协作，从而利用 Elasticsearch 强大的搜索和分析能力。Elastic 的 agent 在内部使用 MCP 来访问 Elasticsearch 工具，体现了 agent 协作（A2A）与工具访问（MCP）之间清晰的分离。
步骤 5：Researcher 使用 MCP 服务器
Researcher agent 访问多个 MCP 服务器来收集信息：
`
1. # Researcher Agent using MCP to access tools
2. async def gather_facts(self, queries):
3. results = []
5. # Use News API MCP Server
6. news_data = await self.mcp_client.invoke_tool(
7. server="news_api_mcp",
8. tool="search_articles",
9. parameters={
10. "query": queries[0],
11. "date_range": "2024-01-01 to 2024-09-30",
12. "sources": ["reuters", "bloomberg", "euractiv"]
13. }
14. )
16. # Use Statistics MCP Server
17. stats_data = await self.mcp_client.invoke_tool(
18. server="eurostat_mcp",
19. tool="get_energy_statistics",
20. parameters={
21. "metrics": ["solar_capacity_mw", "wind_capacity_mw"],
22. "countries": ["EU27"],
23. "year": 2024
24. }
25. )
27. # Use Fact-Checking MCP Server
28. verified_facts = await self.mcp_client.invoke_tool(
29. server="factcheck_mcp",
30. tool="verify_claims",
31. parameters={
32. "claims": [
33. "EU solar capacity increased 40% in 2024",
34. "Germany leads in wind installations"
35. ]
36. }
37. )
39. return self._synthesize_research(news_data, stats_data, verified_facts)
`AI写代码
步骤 6：Researcher 将数据返回给 Reporter
Researcher agent 通过 A2A 将完整的调研结果发回：
`
1. {
2. "message_type": "task_response",
3. "sender": "researcher_agent",
4. "receiver": "reporter_agent",
5. "payload": {
6. "task_id": "research_eu_renewable_2024",
7. "status": "completed",
8. "research_data": {
9. "key_facts": [
10. {
11. "claim": "EU solar capacity reached 260 GW in 2024",
12. "source": "Eurostat Q3 2024 Report",
13. "confidence": 0.95,
14. "verified": true
15. },
16. {
17. "claim": "Germany installed 12 GW of wind capacity in 2024",
18. "source": "German Federal Network Agency",
19. "confidence": 0.92,
20. "verified": true
21. }
22. ],
23. "statistics": {
24. "solar_growth_rate": "35%",
25. "wind_growth_rate": "28%"
26. },
27. "sources_count": 15
28. }
29. }
30. }
`AI写代码
步骤 7：Reporter 撰写文章
Reporter agent 使用调研数据和自身的 LLM 能力来撰写文章。在写作过程中，Reporter 使用 MCP 服务器来获取风格和模板：
`
1. # Reporter Agent writing with MCP assistance
2. async def write_article(self, research_data, assignment):
3. # Get style guidelines via MCP
4. style_guide = await self.mcp_client.get_resource(
5. server="newsroom_mcp",
6. resource="style://editorial/ap_style_guide"
7. )
9. # Get article template via MCP
10. template = await self.mcp_client.get_resource(
11. server="newsroom_mcp",
12. resource="template://articles/news_story"
13. )
15. # Generate article using LLM + research + style
16. draft = await self.llm.generate(
17. prompt=f"""
18. Write a news article following these guidelines:
19. {style_guide}
21. Using this template:
22. {template}
24. Based on this research:
25. {research_data}
27. Assignment: {assignment}
28. """
29. )
31. # Self-evaluate confidence in claims
32. confidence_check = await self._evaluate_confidence(draft)
34. return draft, confidence_check
`AI写代码
步骤 8：低置信度触发重新调研
Reporter agent 评估其草稿后发现有一个观点置信度较低。它向 Researcher agent 发送另一个请求：
`
1. {
2. "message_type": "collaboration_request",
3. "sender": "reporter_agent",
4. "receiver": "researcher_agent",
5. "payload": {
6. "request_type": "fact_verification",
7. "claims": [
8. {
9. "text": "France's nuclear phase-down contributed to 15% increase in renewable capacity",
10. "context": "Discussing policy drivers for renewable growth",
11. "current_confidence": 0.45,
12. "required_confidence": 0.80
13. }
14. ],
15. "urgency": "high"
16. }
17. }
`AI写代码
Researcher 使用 fact-checking MCP 服务器验证该观点，并返回更新的信息：
`
1. {
2. "message_type": "collaboration_response",
3. "sender": "researcher_agent",
4. "receiver": "reporter_agent",
5. "payload": {
6. "verified_claims": [
7. {
8. "original_claim": "France's nuclear phase-down contributed to 15% increase...",
9. "verified_claim": "France's renewable capacity increased 18% in 2024, partially offsetting reduced nuclear output",
10. "confidence": 0.88,
11. "corrections": "Percentage was 18%, not 15%; nuclear phase-down is gradual, not primary driver",
12. "sources": ["RTE France", "French Energy Ministry Report 2024"]
13. }
14. ]
15. }
16. }
`AI写代码
步骤 9：Reporter 修改并提交给 Editor
Reporter 将已验证的事实纳入文章，并通过 A2A 将完成的草稿发送给 Editor agent：
`
1. {
2. "message_type": "task_request",
3. "sender": "reporter_agent",
4. "receiver": "editor_agent",
5. "payload": {
6. "task_id": "edit_renewable_story",
7. "parent_task_id": "story_renewable_energy_2024",
8. "content": {
9. "headline": "Europe's Renewable Revolution: Solar and Wind Surge 30% in 2024",
10. "body": "[Full article text...]",
11. "word_count": 1185,
12. "sources": [/* array of sources */]
13. },
14. "editing_requirements": {
15. "check_style": true,
16. "check_facts": true,
17. "check_seo": true
18. }
19. }
20. }
`AI写代码
步骤 10：Editor 使用 MCP 工具进行审核
Editor agent 使用多个 MCP 服务器来审核文章：
`
1. # Editor Agent using MCP for quality checks
2. async def review_article(self, content):
3. # Grammar and style check
4. grammar_issues = await self.mcp_client.invoke_tool(
5. server="grammarly_mcp",
6. tool="check_document",
7. parameters={"text": content["body"]}
8. )
10. # SEO optimization check
11. seo_analysis = await self.mcp_client.invoke_tool(
12. server="seo_mcp",
13. tool="analyze_content",
14. parameters={
15. "headline": content["headline"],
16. "body": content["body"],
17. "target_keywords": ["renewable energy", "Europe", "solar", "wind"]
18. }
19. )
21. # Plagiarism check
22. originality = await self.mcp_client.invoke_tool(
23. server="plagiarism_mcp",
24. tool="check_originality",
25. parameters={"text": content["body"]}
26. )
28. # Generate editorial feedback
29. feedback = await self._generate_feedback(
30. grammar_issues,
31. seo_analysis,
32. originality
33. )
35. return feedback
`AI写代码
Editor 批准文章并将其发送出去：
`
1. {
2. "message_type": "task_response",
3. "sender": "editor_agent",
4. "receiver": "reporter_agent",
5. "payload": {
6. "status": "approved",
7. "quality_score": 9.2,
8. "minor_edits": [
9. "Changed 'surge' to 'increased' in paragraph 3 for AP style consistency",
10. "Added Oxford comma in list of countries"
11. ],
12. "approved_content": "[Final edited article]"
13. }
14. }
`AI写代码
步骤 11：Publisher 通过 CI/CD 发布
最后，Printer agent 使用 MCP 服务器操作 CMS 和 CI/CD 流水线来发布已批准的文章：
`
1. # Publisher Agent publishing via MCP
2. async def publish_article(self, content, metadata):
3. # Upload to CMS via MCP
4. cms_result = await self.mcp_client.invoke_tool(
5. server="wordpress_mcp",
6. tool="create_post",
7. parameters={
8. "title": content["headline"],
9. "body": content["body"],
10. "status": "draft",
11. "categories": metadata["categories"],
12. "tags": metadata["tags"],
13. "featured_image_url": metadata["image_url"]
14. }
15. )
17. post_id = cms_result["post_id"]
19. # Trigger CI/CD deployment via MCP
20. deploy_result = await self.mcp_client.invoke_tool(
21. server="cicd_mcp",
22. tool="trigger_deployment",
23. parameters={
24. "pipeline": "publish_article",
25. "environment": "production",
26. "post_id": post_id,
27. "schedule": "immediate"
28. }
29. )
31. # Track analytics
32. await self.mcp_client.invoke_tool(
33. server="analytics_mcp",
34. tool="register_publication",
35. parameters={
36. "post_id": post_id,
37. "publish_time": datetime.now().isoformat(),
38. "story_id": metadata["story_id"]
39. }
40. )
42. return {
43. "status": "published",
44. "post_id": post_id,
45. "url": f"https://newsroom.example.com/articles/{post_id}",
46. "deployment_id": deploy_result["deployment_id"]
47. }
`AI写代码
Publisher 通过 A2A 确认文章已发布：
`
1. {
2. "message_type": "task_complete",
3. "sender": "printer_agent",
4. "receiver": "news_chief",
5. "payload": {
6. "task_id": "story_renewable_energy_2024",
7. "status": "published",
8. "publication": {
9. "url": "https://newsroom.example.com/articles/renewable-europe-2024",
10. "published_at": "2025-09-30T17:45:00Z",
11. "post_id": "12345"
12. },
13. "workflow_metrics": {
14. "total_time_minutes": 45,
15. "agents_involved": ["reporter", "researcher", "archive", "editor", "printer"],
16. "iterations": 2,
17. "mcp_calls": 12
18. }
19. }
20. }
`AI写代码
这是配套代码仓库中使用上述相同 agent 的完整 A2A 工作流序列。
| # | From | To | Action | Protocol | Description |
|---|---|---|---|---|---|
| 1 | User | News Chief | Assign Story | HTTP POST | User submits story topic and angle |
| 2 | News Chief | Internal | Create Story | - | Creates story record with unique ID |
| 3 | News Chief | Reporter | Delegate Assignment | A2A | Sends story assignment via A2A protocol |
| 4 | Reporter | Internal | Accept Assignment | - | Stores assignment internally |
| 5 | Reporter | MCP Server | Generate Outline | MCP/HTTP | Creates article outline and research questions |
| 6a | Reporter | Researcher | Request Research | A2A | Sends questions (parallel with 6b) |
| 6b | Reporter | Archivist | Search Archive | A2A JSONRPC | Searches historical articles (parallel with 6a) |
| 7 | Researcher | MCP Server | Research Questions | MCP/HTTP | Uses Anthropic via MCP to answer questions |
| 8 | Researcher | Reporter | Return Research | A2A | Returns research answers |
| 9 | Archivist | Elasticsearch | Search Index | ES REST API | Queries news_archive index |
| 10 | Archivist | Reporter | Return Archive | A2A JSONRPC | Returns historical search results |
| 11 | Reporter | MCP Server | Generate Article | MCP/HTTP | Creates article with research/archive context |
| 12 | Reporter | Internal | Store Draft | - | Saves draft internally |
| 13 | Reporter | News Chief | Submit Draft | A2A | Submits completed draft |
| 14 | News Chief | Internal | Update Story | - | Stores draft, updates status to "draft_submitted" |
| 15 | News Chief | Editor | Review Draft | A2A | Auto-routes to Editor for review |
| 16 | Editor | MCP Server | Review Article | MCP/HTTP | Analyzes content using Anthropic via MCP |
| 17 | Editor | News Chief | Return Review | A2A | Sends editorial feedback and suggestions |
| 18 | News Chief | Internal | Store Review | - | Stores editor feedback |
| 19 | News Chief | Reporter | Apply Edits | A2A | Routes review feedback to Reporter |
| 20 | Reporter | MCP Server | Apply Edits | MCP/HTTP | Revises article based on feedback |
| 21 | Reporter | Internal | Update Draft | - | Updates draft with revisions |
| 22 | Reporter | News Chief | Return Revised | A2A | Returns revised article |
| 23 | News Chief | Internal | Update Story | - | Stores revised draft, status to "revised" |
| 24 | News Chief | Publisher | Publish Article | A2A | Auto-routes to Publisher |
| 25 | Publisher | MCP Server | Generate Tags | MCP/HTTP | Creates tags and categories |
| 26 | Publisher | Elasticsearch | Index Article | ES REST API | Indexes article to news_archive index |
| 27 | Publisher | Filesystem | Save Markdown | File I/O | Saves article as .md file in /articles |
| 28 | Publisher | News Chief | Confirm Publication | A2A | Returns success status |
| 29 | News Chief | Internal | Update Story | - | Updates story status to "published" |
结论
在现代增强型 LLM 基础设施范式中，A2A 和 MCP 都扮演着重要角色。A2A 为复杂的多 agent 系统提供灵活性，但可能可移植性较低且运维复杂度更高。MCP 提供了标准化的工具集成方法，更易于实现和维护，但并非设计用于多 agent 协同。
选择并非二选一。如我们新闻室示例所示，最复杂且高效的 LLM 支持系统通常结合两种方法：agent 通过 A2A 协议进行协调和专业化，同时通过 MCP 服务器访问其工具和资源。这种混合架构在提供多 agent 系统组织效益的同时，也带来了 MCP 的标准化和生态优势。这表明，可能根本不需要选择：直接将两者作为标准方法即可。
作为开发者或架构师，你需要测试并确定两种解决方案的最佳组合，以针对你的具体用例产生最佳结果。理解每种方法的优势、局限性及适用场景，将帮助你构建更高效、可维护且可扩展的 AI 系统。
无论你是在构建数字新闻室、客户服务平台、研究助理还是其他 LLM 驱动的应用，仔细考虑你的协调需求（A2A）和工具访问需求（MCP）都将为成功奠定基础。
附加资源
- Elasticsearch Agent Builder: www.elastic.co/docs/soluti…
- A2A Specification: a2a-protocol.org/latest/spec…
- A2A and MCP Integration: a2a-protocol.org/latest/topi…
- Model Context Protocol: modelcontextprotocol.io
