ScaleMCP - A dynamic MCP search tool launched by PwC
ScaleMCP is a tool selection methodology developed by PwC that dynamically equips Large Language Model (LLM) agents with Model Context Protocol (MCP) tools. It is based on an automatically synchronized tool storage system and MCP server...
What is ScaleMCP?
ScaleMCP, a tool selection methodology developed by PwC, dynamically equips Large Language Model (LLM) agents with Model Context Protocol (MCP) tools. Based on an automatic synchronization system between the tool storage and the MCP server, it addresses the inefficiencies and inconsistencies caused by existing frameworks relying on manual updates of local tool repositories. At its core, ScaleMCP automatically synchronizes the tool index pipeline, ensuring consistency between the tool storage and the MCP server based on CRUD operations (create, read, update, delete). ScaleMCP introduces a Tool Document Weighted Average (TDWA) embedding strategy, which selectively emphasizes key sections of tool documents, improving tool retrieval and agent invocation performance.
ScaleMCP's main functions
- Dynamic tool discovery and configurationLLM Agents dynamically discover and load the required MCP tools during multi-round interactions without prior configuration.
- Automatic synchronization tool storage systemBased on CRUD operations (create, read, update, delete), it keeps synchronized with the MCP server to ensure the real-time updates and consistency of the tool's storage system.
- Supports multiple retrieval and embedding modelsIt is compatible with various LLM models, embedding models, and retrieval types, and has good scalability and flexibility.
- Improve tool usage and task completion ratesImprove the performance of LLM Agents in complex tasks, especially in multi-hop tool invocation scenarios.
The technical principles of ScaleMCP
- Model Context Protocol (MCP)MCP is a standardized protocol used to connect LLMs with external tools, data sources, and hints. It allows developers to expose tools or APIs based on the MCP server, simplifying the process of AI systems accessing necessary data. ScaleMCP uses the MCP server as the single source of truth for tools, keeping CRUD operations synchronized with the tool storage system.
- Automatic synchronization tool storage systemScaleMCP is based on an automatic synchronization tool index pipeline. It periodically retrieves a list of tools from the MCP server and calculates the SHA-256 hash of each tool. The newly retrieved tool hash is compared with existing hashes in the storage system. If a mismatch is found, the tool index in the storage system is updated. ScaleMCP supports various storage systems, such as vector databases and graph databases, and selects the appropriate storage method based on the tool's dependencies.
- Tool Document Weighted Average (TDWA) Embedding StrategyTraditional tool document embedding methods (such as simple concatenation or averaging) cannot effectively distinguish the importance of different parts of the tool document. TDWA assigns weights to each part of the tool document (such as tool name, description, parameters, composition problem, etc.) and calculates a weighted average embedding vector, which more accurately represents the semantic information of the tool.
- Dynamic tool retrieval and invocation via LLM proxiesScaleMCP provides LLM Agents with an MCP retrieval tool. Agents search for relevant MCP tools based on keywords and load them into their context. LLM Agents dynamically invoke multiple tools based on user queries, obtain tool responses, perform inference, and ultimately generate an answer. During multiple rounds of interaction, the Agent can re-query the tool storage system and dynamically adjust tool selection and invocation strategies.
ScaleMCP project address
- arXiv technical paper:https://arxiv.org/pdf/2505.06416
Application scenarios of ScaleMCP
- Financial data analysisDynamically load financial APIs, update data in real time, and quickly respond to complex financial queries.
- Intelligent Customer ServiceIntegrate enterprise APIs, call tools in real time, and provide precise customer support.
- Medical and health consultationIt connects to medical databases and appointment systems, supporting disease inquiry and appointment services.
- Educational guidanceIt integrates with the school's information system, dynamically calls up tools, and allows users to query course information and exam results.
- Smart OfficeIt connects to file management and scheduling systems, supporting file retrieval and meeting arrangement.