← Back to directory
H

Hyper MCP

Community
A fast, secure MCP server that extends its capabilities through WebAssembly plugins.
GitHub source repository ↗
★ 880 Stars Category · Dev Tools Very popular Source revision fb2b24fbbe66
67FMRS · C
Reliability
11/20
Security and permissions
13/20
Maintenance
15/20
Documentation
16/20
Setup experience
12/20

hyper-mcp is an innovative MCP server that uses WebAssembly plugins for high extensibility. It is secure, lightweight, and supports multiple clients. The main limitations are stdio-only transport and the requirement for cosign to load OCI plugins. Overall, it is well-suited for developers who want to customize and expand AI tooling.

Read the FMRS scoring method →

hyper-mcp is a fast and secure MCP server that extends its capabilities through WebAssembly plugins. It allows writing plugins in any language that compiles to WebAssembly, and distributing them via standard OCI registries (like Docker Hub). Built on Extism for robust plugin support, it provides sandboxing to limit network, filesystem, and memory access. It supports the stdio transport protocol, is lightweight enough for resource-constrained environments, and can be deployed anywhere: serverless, edge, mobile, IoT. Dynamic plugin loading/unloading and tool name prefixing are supported to prevent collisions.

Tools

hyper_mcp-list_plugins
Lists all loaded plugins.
hyper_mcp-load_plugin
Dynamically loads a plugin at runtime.
hyper_mcp-unload_plugin
Dynamically unloads a plugin at runtime.

Setup

  1. Install prerequisites: cosign (required for loading plugins from OCI registries with signature verification). 2. Install hyper-mcp via Homebrew (brew install hyper-mcp or brew install hyper-mcp-rs/tap/hyper-mcp), via cargo (cargo install hyper-mcp), or download a pre-built binary from GitHub Releases. 3. Create a config file (e.g., ~/.config/hyper-mcp/config.json) specifying the WASM plugins to load. 4. Start the server from the command line: hyper-mcp.
claude_desktop_config.json
{
  "mcpServers": {
    "hyper-mcp": {
      "command": "/path/to/hyper-mcp"
    }
  }
}

Fit and risk

Best for

  • Developers who need highly extensible and secure AI services with the flexibility to write plugins in multiple languages.
  • Teams that want to distribute and manage plugins via OCI registries.
  • Enterprise applications requiring sandboxing and resource limits.

Not for

  • Clients that require HTTP transports (like SSE or streamable-http), since hyper-mcp defaults to stdio only.
  • Users who prefer not to install cosign and still load OCI plugins (unless they set insecure_skip_signature).
  • Users who need specific built-in tools (e.g., database operations) without wanting to configure plugins themselves.

Required permissions

  • Read config files (for plugin configuration).
  • Execute cosign CLI to verify OCI plugin signatures.
  • Network access (to download plugins or for plugin runtime, e.g., allowed hosts).
  • Filesystem access (for local file plugins or logs).
  • Environment variables (e.g., RUST_LOG, HYPER_MCP_LOG_PATH, HYPER_MCP_INSECURE_SKIP_SIGNATURE).

Risks and side effects

  • Loading malicious or unverified plugins could compromise the system or leak data, though signature verification and sandboxing mitigate risks.
  • OCI signature verification depends on cosign; skipping it reduces security.
  • Misconfiguring plugins may allow them to access unauthorized network or filesystem resources.

Troubleshooting

  1. Ensure cosign is installed and in PATH, otherwise loading OCI plugins will fail.
  2. Check that the config file path is correct and the JSON is valid.
  3. Use `RUST_LOG=debug hyper-mcp` to get verbose logs for troubleshooting.
  4. If log file location is not as expected, verify the HYPER_MCP_LOG_PATH environment variable.

Use cases

Add AI capabilities to applications such as Claude Desktop, Cursor IDE, and other MCP-compatible apps.
Extend the MCP server with custom WebAssembly plugins for tasks like time queries, QR code generation, hashing, IP lookup, web fetching, and more.
Deploy lightweight AI services in resource-constrained environments.

Supported clients

Claude DesktopFull support
Cursor IDEFull support
Claude CodeFull support