MCP Aggregator Server

MCP Aggregator Server

Provides a unified MCP interface that proxies requests to multiple backend servers including memory/knowledge graph and vector database services. Enables seamless access to distributed MCP tools through a single endpoint with automatic routing, health monitoring, and retry logic.

Category
访问服务器

README

MCP Aggregator Server

Unified MCP interface that proxies requests to multiple backend MCP servers.

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     MCP Client                              │
│              (Claude, IDE, etc.)                            │
└────────────────────┬────────────────────────────────────────┘
                     │
                     │ Connect to single endpoint
                     ▼
┌─────────────────────────────────────────────────────────────┐
│         Aggregator MCP Server (Port 8003)                   │
│  ┌──────────────────────────────────────────────────────┐   │
│  │  Unified MCP Interface                               │   │
│  │  - 19 tools total (2 health + 10 memory + 7 vector) │   │
│  │  - Handles routing internally                        │   │
│  │  - Single /mcp/sse & /mcp/messages endpoint          │   │
│  └──────────────────────────────────────────────────────┘   │
└────────┬──────────────────────────────────────────────────┬──┘
         │                                                  │
         │ HTTP Proxy                                       │ HTTP Proxy
         ▼                                                  ▼
┌──────────────────────┐                        ┌──────────────────────┐
│  ZepAI Memory Server │                        │  LTM Vector Server   │
│  (Port 8002)         │                        │  (Port 8000)         │
│                      │                        │                      │
│ - Knowledge Graph    │                        │ - Vector Database    │
│ - Conversation Memory│                        │ - Code Indexing      │
│ - 10 tools           │                        │ - 7 tools            │
└──────────────────────┘                        └──────────────────────┘

Features

  • Unified Interface: Single MCP endpoint for all connected servers
  • Transparent Proxying: Automatically routes requests to appropriate backend servers
  • Health Monitoring: Built-in health checks for all connected servers
  • Retry Logic: Automatic retry with exponential backoff for failed requests
  • Error Handling: Comprehensive error handling and logging
  • Extensible: Easy to add new backend servers

Installation

  1. Install dependencies:
pip install -r requirements.txt
  1. Configure environment (edit .env):
# Aggregator Server
AGGREGATOR_HOST=0.0.0.0
AGGREGATOR_PORT=8003

# Memory Server (FastMCP Server)
MEMORY_SERVER_URL=http://localhost:8002
MEMORY_SERVER_TIMEOUT=30

# Graph Server (for future use)
GRAPH_SERVER_URL=http://localhost:8000
GRAPH_SERVER_TIMEOUT=30

Running

Start all servers in order:

Terminal 1 - LTM Vector Server (Port 8000):

cd LTM
python mcp_server/server_streamable_http.py

Terminal 2 - ZepAI FastMCP Server (Port 8002):

cd ZepAI/fastmcp_server
python server_http.py

Note: This automatically loads the Memory Layer and exposes both FastAPI + MCP on port 8002

Terminal 3 - MCP Aggregator (Port 8003):

cd mcp-aggregator
python aggregator_server.py

See START_SERVERS.md for detailed startup guide.

Available Tools

Health & Status

  • health_check() - Check health of all connected servers
  • get_server_info() - Get information about connected servers

Memory Server Tools (Port 8002)

Search

  • memory_search(query, project_id, limit, use_llm_classification) - Search knowledge graph
  • memory_search_code(query, project_id, limit) - Search code memories

Ingest

  • memory_ingest_text(text, project_id, metadata) - Ingest plain text
  • memory_ingest_code(code, language, project_id, metadata) - Ingest code
  • memory_ingest_json(data, project_id, metadata) - Ingest JSON data
  • memory_ingest_conversation(conversation, project_id) - Ingest conversation

Admin

  • memory_get_stats(project_id) - Get project statistics
  • memory_get_cache_stats() - Get cache statistics

LTM Vector Server Tools (Port 8000)

Repository Processing

  • ltm_process_repo(repo_path) - Process repository for vector indexing

Vector Search

  • ltm_query_vector(query, top_k) - Query vector database for semantic code search
  • ltm_search_file(filepath) - Search for specific file in vector database

File Management

  • ltm_add_file(filepath) - Add file to vector database
  • ltm_delete_by_filepath(filepath) - Delete file from vector database
  • ltm_delete_by_uuids(uuids) - Delete vectors by UUIDs

Code Analysis

  • ltm_chunk_file(file_path) - Chunk file using AST-based chunking

Testing

1. Check Server Health

curl http://localhost:8003/mcp/sse

2. Access OpenAPI Docs

http://localhost:8003/docs

3. Test a Tool via MCP

# Using MCP client
mcp-client http://localhost:8003/mcp health_check

Configuration

Environment Variables

Variable Default Description
AGGREGATOR_HOST 0.0.0.0 Aggregator server host
AGGREGATOR_PORT 8003 Aggregator server port
MEMORY_SERVER_URL http://localhost:8002 Memory server URL
MEMORY_SERVER_TIMEOUT 30 Memory server timeout (seconds)
GRAPH_SERVER_URL http://localhost:8000 Graph server URL
GRAPH_SERVER_TIMEOUT 30 Graph server timeout (seconds)
LOG_LEVEL INFO Logging level
MAX_RETRIES 3 Max retries for failed requests
RETRY_DELAY 1 Delay between retries (seconds)
HEALTH_CHECK_INTERVAL 30 Health check interval (seconds)

Adding New Backend Servers

To add a new backend server (e.g., Graph Server):

  1. Update config.py:
GRAPH_SERVER_URL = os.getenv("GRAPH_SERVER_URL", "http://localhost:8000")
GRAPH_SERVER_TIMEOUT = int(os.getenv("GRAPH_SERVER_TIMEOUT", "30"))
  1. Update mcp_client.py:
class AggregatorClients:
    def __init__(self):
        # ... existing clients ...
        self.graph_client = MCPServerClient(
            "Graph Server",
            config.GRAPH_SERVER_URL,
            config.GRAPH_SERVER_TIMEOUT
        )
  1. Add tools in aggregator_server.py:
@mcp.tool()
async def graph_query(cypher: str) -> Dict[str, Any]:
    """Query Neo4j graph database"""
    clients = await get_clients()
    return await clients.graph_client.proxy_request(
        "POST",
        "/query",
        json_data={"cypher": cypher},
        retries=config.MAX_RETRIES
    )

Troubleshooting

Connection Refused

  • Ensure all backend servers are running
  • Check URLs in .env file
  • Verify ports are not blocked by firewall

Timeout Errors

  • Increase MEMORY_SERVER_TIMEOUT or GRAPH_SERVER_TIMEOUT in .env
  • Check backend server performance
  • Verify network connectivity

Health Check Failing

  • Run health_check() tool to diagnose
  • Check backend server logs
  • Verify backend servers are responding

Development

Project Structure

mcp_aggregator/
├── aggregator_server.py    # Main MCP server
├── config.py               # Configuration management
├── mcp_client.py           # HTTP clients for backend servers
├── requirements.txt        # Python dependencies
├── .env                    # Environment variables
├── __init__.py             # Package initialization
└── README.md               # This file

Adding Logging

import logging
logger = logging.getLogger(__name__)
logger.info("Message")
logger.error("Error")

Future Enhancements

  • [ ] Add Graph/Vector DB server integration
  • [ ] Implement caching layer
  • [ ] Add request rate limiting
  • [ ] Implement server load balancing
  • [ ] Add metrics/monitoring
  • [ ] Support for server discovery
  • [ ] WebSocket support for real-time updates

License

Same as parent project (Innocody)

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选