Wassette, a Microsoft-developed, open-source Model Context Protocol (MCP) server, provides a secure environment for AI agents to execute untrusted tools. Built with Rust and utilizing the WebAssembly (Wasm) Component Model and Wasmtime runtime, it ensures safe operation. Distributed as a standalone binary, Wassette connects to various AI agents and Large Language Models (LLMs), acting as a trusted execution environment. It enables dynamic discovery and invocation of Wasm Component functions, enhancing AI capabilities with controlled permissions.

What does Wassette connect to?

Wassette functions as a Model Context Protocol (MCP) server, linking directly to AI agents and Large Language Models (LLMs). Its primary role is to provide a trusted execution environment for tools that might otherwise be untrusted. It supports communication via stdio in its traditional MCP server mode. Additionally, it can host Wasm Component-based MCP tools over a network, expanding its reach. Confirmed compatible AI clients include Visual Studio Code (with GitHub Copilot), Cursor, Claude Code, and Gemini CLI, allowing a broad range of AI interactions.

Core features and tools

Wassette exposes the typed library interfaces of loaded WebAssembly Components as MCP tools, offering dynamic functionality rather than a fixed set of built-in tools. AI agents can discover and invoke specific functions within these hosted Wasm Components. Examples include a time component for current time, a reverse_text tool for string manipulation, and a fetch component for HTTP requests. The wassette command-line interface (CLI) provides commands such as "wassette component load", "wassette permission grant", "wassette policy get", and "wassette serve" for server and component management.

Who is it for?

Wassette is designed for Application Developers seeking to simplify tool creation and deployment, DevOps Engineers managing secure environments, and developers building or extending AI agents. It supports workflows where AI agents need to securely extend their capabilities by executing external components in a sandboxed manner. Organizations can also use it to distribute and manage cryptographically signed AI tools. Skip this if your primary need is user authentication for an MCP server, as Wassette focuses on tool capability security.

How to get started

No credentials are required for server access, as the authentication model focuses on capability-based security for the tools running within Wassette. Communication can occur over stdio as a transport protocol, particularly in "traditional mode" for MCP servers.

Use cases

• Securely extending AI agents: AI agents fetch WebAssembly Components from OCI registries, executed by Wassette in a sandboxed environment with controlled permissions for new tasks.
• Developing and testing WebAssembly Components: Developers build Wasm Components, load them locally into Wassette, define permissions, and test functionality with MCP-compatible AI agents.
• Distributing and managing secure AI tools: Organizations publish cryptographically signed WebAssembly Components to OCI registries, enabling AI agents to discover and securely execute these tools.

Wassette offers a secure, open-source platform for AI agents to run WebAssembly Components, enhancing their capabilities within a controlled environment. Its design prioritizes capability-based security for tools, using fine-grained permissions and cryptographic verification for component distribution. This Microsoft tool supports various AI clients and development workflows, making it a valuable addition for extending AI agent functionality. Start by loading a "time component" to see how an AI agent can fetch the current time.