MCPStore

MCPStore is an enterprise-grade MCP tool management solution and server built with Rust that simplifies AI agent tool integration, service lifecycle management, and system monitoring. It functions using a Kubernetes-style control plane and data plane architecture, allowing developers and system administrators to orchestrate tools centrally while executing invocations with minimal resource overhead. MCPStore connects diverse Model Context Protocol services, such as local command-line tools and remote web endpoints, exposing them unified across multiple agent environments. Software engineers building AI workflows use MCPStore to register, monitor, restart, and scope tools per agent, preventing configuration conflicts and reducing maintenance. It integrates natively with popular LLM agent frameworks, including LangChain, LangGraph, OpenAI, AutoGen, CrewAI, and LlamaIndex. By exposing both a command-line interface, a local web dashboard, desktop applications, and an HTTP API, MCPStore provides complete observability and dynamic control over all operational MCP tools across an organization's AI infrastructure.

Category: AI & LLM Tooling

Tags: enterprise, integration, management, monitoring

Visit MCPStore

How to install and configure MCPStore

To set up and run MCPStore, follow these steps: 1. Install the CLI using npm or the installation script: bash npm install -g ip2a@mcpstore or bash curl -fsSL https://raw.githubusercontent.com/ip2a/mcpstore/main/install.sh | bash 2. Add your target MCP service using the CLI: bash mcpstore add wiki https://example.com/mcp 3. Connect and verify the service status: bash mcpstore connect wiki mcpstore list 4. If using Python or LangChain workflows, install the SDK via pip install mcpstore and import MCPStore to initialize the store and convert tools directly into your framework. 5. Optionally expose the local HTTP API or launch the web dashboard using mcpstore web or mcpstore api --host 127.0.0.1 --port 1820. For more client configuration details, consult the mcpstore repository.

What you can do with MCPStore

• Exposing remote or local MCP services to agent frameworks like LangChain, AutoGen, and CrewAI through a centralized management layer. - Isolating tool access per agent using scoped agent identifiers to avoid polluting prompts with irrelevant external functions. - Managing the lifecycle of multiple MCP services through commands for adding, updating, restarting, disconnecting, and monitoring runtime status. - Operating a low-footprint data plane on resource-constrained host machines while centralizing tool orchestration through an external control plane. - Providing a local desktop interface or web UI for developers to inspect service configuration schemas and debug tool invocations.

Key facts

• https://github.com/whillhill/mcpstore
• AI & LLM Tooling, Developer Tools & Code Intelligence, Monitoring & Observability
• enterprise, integration, management, monitoring

Part of MCP Servers

How do I install MCPStore?

You can install MCPStore globally through npm with the command npm install -g ip2a@mcpstore or by running the installation script via curl. For Python-based agent development, install the library using pip install mcpstore. Pre-built desktop applications for macOS, Windows, and Linux are also provided in the GitHub releases.

What can MCPStore do?

MCPStore provides orchestration, lifecycle management, and monitoring for Model Context Protocol servers. It allows you to register tools, inspect parameters, run local web dashboards, group MCP services by specific agent identifiers, and export active tools to frameworks like LangChain, CrewAI, AutoGen, and LlamaIndex.

Which agent frameworks work with MCPStore?

MCPStore supports multiple major LLM frameworks via its Python SDK. It provides helper methods to export tools into LangChain, LangGraph, standard OpenAI tool formats, AutoGen, CrewAI, and LlamaIndex, allowing unified tool access across different runtime setups.

What is the control plane and data plane architecture in MCPStore?

MCPStore splits operations into a control plane and a data plane. The control plane orchestrates, updates, and monitors tool definitions. The data plane handles the actual tool invocations. This enables running only the data plane on resource-limited hosts without maintaining local server configurations.