GEO
AI大模型

如何用SmartBuckets+MCP快速构建AI代理？2026年开发效率提升指南

2026/3/27
如何用SmartBuckets+MCP快速构建AI代理？2026年开发效率提升指南
AI Summary (BLUF)

LiquidMetal's SmartBuckets technology combined with Anthropic's Model Context Protocol (MCP) reduces AI agent development time from months to days by eliminating RAG pipeline bottlenecks and providing automatic knowledge graph creation.

原文翻译： LiquidMetal的SmartBuckets技术与Anthropic的模型上下文协议（MCP）相结合，通过消除RAG管道瓶颈并提供自动知识图谱创建，将AI代理开发时间从数月缩短至数天。

Introduction

Hey HN, we've tackled one of the most persistent challenges in AI agent development: the RAG (Retrieval-Augmented Generation) pipeline bottleneck. At LiquidMetal, we've integrated our proprietary SmartBuckets technology with Anthropic's Model Context Protocol (MCP). This combination has dramatically accelerated development cycles, reducing the time required to build knowledge-powered agents from several months down to just days.

各位 HN 的读者，我们解决了一个在构建 AI 智能体时最令人头疼的问题：RAG（检索增强生成）流程的瓶颈。在 LiquidMetal，我们将我们专有的 SmartBuckets 技术与 Anthropic 的模型上下文协议（MCP）相结合。这一组合极大地加速了开发周期，将构建知识驱动型智能体所需的时间从数月缩短至数天。

The Core Problem: The RAG Bottleneck

Constructing AI agents capable of leveraging organizational knowledge has traditionally been a massive engineering undertaking, typically requiring six months or more of dedicated work. Development teams invest countless hours into building and fine-tuning a complex suite of components.

构建能够利用组织知识的 AI 智能体，传统上是一项庞大的工程任务，通常需要六个月或更长时间的专门工作。开发团队需要投入大量时间来构建和微调一系列复杂的组件。

This effort is distributed across several critical areas:

  • Document Processing Pipelines: Creating robust systems to ingest and normalize diverse file formats.
  • Chunking Strategies: Developing intelligent methods to split documents into semantically meaningful segments for retrieval.
  • Embedding Generation: Implementing and optimizing models to convert text into numerical vectors.
  • Entity Extraction & Knowledge Graph Creation: Building systems to identify key concepts and their relationships within the data.
  • Vector Database Configuration: Selecting, deploying, and tuning specialized databases for high-speed similarity search.
  • Retrieval Algorithm Development: Crafting logic to find the most relevant information for a given query.
  • Context Assembly & Management: Designing methods to compile retrieved data into a coherent context window for the LLM.

这些工作分布在几个关键领域：

  • 文档处理流程：创建健壮的系统来摄取和规范化各种文件格式。
  • 分块策略：开发智能方法，将文档分割成具有语义意义的片段以便检索。
  • 嵌入生成：实现并优化模型，将文本转换为数值向量。
  • 实体提取与知识图谱创建：构建系统以识别数据中的关键概念及其相互关系。
  • 向量数据库配置：选择、部署和调优用于高速相似性搜索的专用数据库。
  • 检索算法开发：设计逻辑，为给定查询找到最相关的信息。
  • 上下文组装与管理：设计方法，将检索到的数据汇编成供大语言模型使用的连贯上下文窗口。

Our Solution: SmartBuckets + MCP

SmartBuckets addresses this complexity head-on by providing a fully integrated knowledge engine. It eliminates the need to construct each of these components from the ground up. When combined with Anthropic's Model Context Protocol (MCP), which facilitates direct, structured communication between applications and AI models, it creates a powerful and streamlined development stack.

SmartBuckets 通过提供一个完全集成的知识引擎，直接应对了这种复杂性。它消除了从头构建每个组件的需求。当与 Anthropic 的模型上下文协议（MCP）结合时——该协议促进了应用程序与 AI 模型之间直接、结构化的通信——它创造了一个强大且精简的开发技术栈。

Technical Architecture Overview

The architecture is designed for efficiency and direct integration. SmartBuckets serves as the central knowledge processing and storage layer, while MCP acts as the standardized conduit for AI models to access this enriched data on-demand.

该架构专为效率和直接集成而设计。SmartBuckets 作为核心的知识处理和存储层，而 MCP 则作为标准化通道，让 AI 模型能够按需访问这些增强后的数据。

AI Decomposition: The Foundational Process

When a file is uploaded to a SmartBucket, it initiates an intelligent, multi-stage process we term AI Decomposition. This process is the cornerstone of how SmartBuckets transforms raw, unstructured files into AI-ready, semantically enriched resources.

当文件上传到 SmartBucket 时，会启动一个我们称之为 AI 分解 的智能多阶段过程。这个过程是 SmartBuckets 如何将原始的、非结构化的文件转化为 AI 就绪、语义增强的资源的基础。

The decomposition unfolds through several key stages:

  1. Content Identification & Extraction: The system first analyzes the file to identify and extract heterogeneous content types, including text, images, tables, and metadata.
  2. Specialized AI Processing: Each extracted component is then routed through specialized AI models optimized for its content type (e.g., vision models for images, NLP models for text).
  3. Optimized Storage: The enhanced data is stored in purpose-built datastores (vector stores, graph databases, etc.), meticulously preserving the relationships between different components.
  4. Immediate Query Availability: The entire processed corpus becomes instantly available for querying by AI agents via the MCP interface.

分解过程通过几个关键阶段展开：

  1. 内容识别与提取：系统首先分析文件，识别并提取异构的内容类型，包括文本、图像、表格和元数据。
  2. 专业化 AI 处理：然后，每个提取出的组件被路由到针对其内容类型优化的专用 AI 模型进行处理（例如，图像使用视觉模型，文本使用 NLP 模型）。
  3. 优化存储：增强后的数据存储在专门构建的数据存储（向量数据库、图数据库等）中，并精心保留了不同组件之间的关系。
  4. 即时查询可用性：整个处理后的语料库通过 MCP 接口立即可供 AI 智能体查询。

Automatic Knowledge Graph Creation

A key differentiator for SmartBuckets is its ability to move beyond simple vector search. The system automatically constructs a knowledge graph from uploaded documents, which provides a structured representation of information that significantly enhances retrieval accuracy and reduces model "hallucinations."

SmartBuckets 的一个关键区别在于它能够超越简单的向量搜索。系统会自动从上传的文档中构建一个 知识图谱，这提供了信息的结构化表示，能显著提高检索准确性并减少模型的“幻觉”。

The knowledge graph creation involves:

  1. Automatic Entity & Relationship Extraction: The system identifies key entities (people, places, concepts) and the semantic relationships between them within the documents.
  2. Graph Construction: These entities and relationships are used to build an interconnected graph, visually and logically mapping how pieces of information relate.
  3. Metadata Enrichment: Data points within the graph are tagged with additional metadata, further refining the retrieval process.

知识图谱的创建包括：

  1. 自动实体与关系提取：系统识别文档中的关键实体（人物、地点、概念）以及它们之间的语义关系。
  2. 图谱构建：这些实体和关系被用来构建一个互连的图，从视觉和逻辑上映射信息片段之间的关联。
  3. 元数据丰富化：图中的数据点被附加了额外的元数据标签，进一步优化检索过程。

Integrated AI Models and Data Stores

The processed data from the AI Decomposition pipeline is not stored in a single monolithic database. Instead, it is distributed across multiple specialized storage systems, each chosen for optimal performance for a specific type of query or analysis.

来自 AI 分解流程的处理后数据并非存储在单一的庞大数据仓库中。相反，它分布在多个专门的存储系统中，每个系统都因其对特定类型查询或分析的最佳性能而被选用。

The processing pipeline itself incorporates various analysis models that perform tasks such as:

  • PII (Personally Identifiable Information) Detection: Automatically identifies and can help manage sensitive personal data within documents.
  • Content Safety Screening: (Feature in development) Aims to flag harmful or inappropriate content.

处理流程本身包含了各种分析模型，用于执行以下任务：

  • PII（个人身份信息）检测：自动识别并有助于管理文档中的敏感个人数据。
  • 内容安全筛查：（开发中的功能）旨在标记有害或不适当的内容。

Technical Implementation: A Practical Example

Integrating SmartBuckets with MCP-compatible applications, such as Claude Desktop, is designed to be straightforward, requiring minimal configuration code.

将 SmartBuckets 与 MCP 兼容的应用程序（如 Claude Desktop）集成，其设计目标是简单直接，只需最少的配置代码。

Here is a step-by-step example for attaching a SmartBucket to Claude Desktop:

  1. Navigate within Claude Desktop: Claude → Settings → Developer → Edit Config.
  2. Add a configuration block for the LiquidMetal MCP server. You will need an API key, which can be obtained from your LiquidMetal account dashboard (Settings → API Keys).

以下是将 SmartBucket 连接到 Claude Desktop 的分步示例：

  1. 在 Claude Desktop 中导航至：Claude → 设置 → 开发者 → 编辑配置。
  2. 为 LiquidMetal MCP 服务器添加一个配置块。您需要一个 API 密钥，可以从您的 LiquidMetal 账户仪表板（设置 → API 密钥）获取。

The configuration snippet would look similar to the following (remember to replace <liquidmetal_key_here> with your actual API key):

配置代码片段类似于以下内容（请记住将 <liquidmetal_key_here> 替换为您实际的 API 密钥）：

{
  "mcpServers": {
    "liquidmetal": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.raindrop.run/sse",
        "--header",
        "Authorization: Bearer ${RAINDROP_API_KEY}"
      ],
      "env": {
        "RAINDROP_API_KEY": "<liquidmetal_key_here>"
      }
    }
  }
}

After saving the configuration and restarting Claude Desktop, your processed documents from SmartBuckets become immediately accessible within the conversation interface, enabling the AI to reference your proprietary knowledge base directly.

保存配置并重启 Claude Desktop 后，您在 SmartBuckets 中处理过的文档将立即在对话界面中可用，使 AI 能够直接引用您的专有知识库。

Future Roadmap

We are actively developing new features to expand the capabilities of the SmartBuckets and MCP integration. Our focus areas include:

  • Direct CREATE Operations via MCP: Enabling the creation of new SmartBuckets and content directly through MCP instructions.
  • Expanded File Type Support: Adding robust processing pipelines for video, source code, application logs, and other complex data types.
  • Community-Driven Development: We prioritize features based on user feedback. We encourage you to tell us what capabilities are missing for your use cases.

我们正在积极开发新功能，以扩展 SmartBuckets 和 MCP 集成的能力。我们的重点领域包括：

  • 通过 MCP 的直接 CREATE 操作：支持直接通过 MCP 指令创建新的 SmartBuckets 和内容。
  • 扩展的文件类型支持：为视频、源代码、应用程序日志和其他复杂数据类型添加强大的处理流程。
  • 社区驱动的开发：我们根据用户反馈来确定功能的优先级。我们鼓励您告诉我们，对于您的使用场景还缺少哪些功能。

Conclusion and Call for Feedback

We are excited to release this integration to the Hacker News community. By combining a pre-built, intelligent knowledge engine (SmartBuckets) with a standardized model interface protocol (MCP), we believe we can unlock a new level of productivity in AI agent development.

我们很高兴能将此集成版本发布给 Hacker News 社区。通过将预构建的智能知识引擎（SmartBuckets）与标准化的模型接口协议（MCP）相结合，我们相信我们能够将 AI 智能体开发的生产力提升到一个新的水平。

To get started, visit our documentation at https://docs.liquidmetal.ai/. Use the code HN-MCP-100 during sign-up to receive $100 in free LiquidMetal credits. We eagerly await your experiences, technical questions, and suggestions.

要开始使用，请访问我们的文档网站 https://docs.liquidmetal.ai/。注册时使用代码 HN-MCP-100 即可获得 100 美元的免费 LiquidMetal 积分。我们热切期待您的使用体验、技术问题和建议。

常见问题（FAQ）

SmartBuckets技术如何解决RAG管道瓶颈？

SmartBuckets提供完全集成的知识引擎，自动处理文档分块、嵌入生成、知识图谱创建等复杂组件，无需从零构建，从而消除传统RAG开发的主要瓶颈。

MCP协议在架构中起什么作用？

MCP作为标准化通信通道，让AI模型能直接、结构化地访问SmartBuckets处理后的语义增强数据，实现应用程序与AI模型的高效集成。

AI分解过程具体包含哪些步骤？

包括内容识别提取、专用AI模型处理等阶段，将非结构化文件转化为AI就绪的语义资源，这是SmartBuckets自动知识图谱创建的核心流程。

标签
AI应用开发生成式AI大语言模型AI大模型人工智能
← 返回文章列表
分享到：微博
下一篇
Ragtime是什么？2026年自托管RAG API与MCP服务器深度解析

版权与免责声明：本文仅用于信息分享与交流，不构成任何形式的法律、投资、医疗或其他专业建议，也不构成对任何结果的承诺或保证。

文中提及的商标、品牌、Logo、产品名称及相关图片/素材，其权利归各自合法权利人所有。本站内容可能基于公开资料整理，亦可能使用 AI 辅助生成或润色；我们尽力确保准确与合规，但不保证完整性、时效性与适用性，请读者自行甄别并以官方信息为准。

若本文内容或素材涉嫌侵权、隐私不当或存在错误，请相关权利人/当事人联系本站，我们将及时核实并采取删除、修正或下架等处理措施。 也请勿在评论或联系信息中提交身份证号、手机号、住址等个人敏感信息。