microsoft/wassette

Wassette: A security-oriented runtime that runs WebAssembly Components via MCP

What it solves

Wassette provides a secure way to extend AI agents with new tools. It eliminates the need to leave the chat window to add functionality, while ensuring that these tools run in a protected environment to prevent security risks.

How it works

Wassette acts as a runtime that executes WebAssembly (Wasm) Components. It uses the Model Context Protocol (MCP) to connect these components to AI agents. By leveraging the Wasmtime security sandbox, it provides browser-grade isolation for the tools the agent uses.

Who it’s for

Developers and users of AI agents (such as GitHub Copilot, Claude Code, and Codex CLI) who want to add reusable, sandboxed tools to their agents without compromising system security.

Highlights

  • Secure Isolation: Uses Wasmtime to provide a sandbox for tool execution.
  • MCP Integration: Runs Wasm Components via the Model Context Protocol.
  • Seamless Extension: Allows agents to load components from OCI registries (e.g., oci://ghcr.io/microsoft/time-server-js:latest).
  • Reusable Components: Uses generic Wasm Components that are not tied to any specific protocol.

Related

  • Project
    hyper-mcp-rs/hyper-mcp

    A fast, secure MCP server that uses WebAssembly sandboxing to allow AI applications to safely execute plugins written in any WASM-compatible language.

  • Project
    1mcp-app/agent

    A unified runtime for MCP servers that aggregates multiple servers into one process and provides a progressive discovery workflow to prevent agent context sprawl.

  • Project
    WasmEdge/WasmEdge

    A lightweight, high-performance WebAssembly runtime that provides a secure sandbox for running LLMs, microservices, and plug-ins across edge devices and servers.

  • Project
    xyTom/coding-tools-mcp

    A model-neutral coding runtime using the Model Context Protocol that gives AI agents secure access to read, patch, and execute commands within a codebase.

  • Project
    nakasyou/local-mcp

    A Model Context Protocol (MCP) server that exposes local filesystem and command execution capabilities to AI agents with built-in sandboxing and user approval workflows.