stacklok/toolhive

What it solves

ToolHive provides a secure, scalable way to run and manage Model Context Protocol (MCP) servers. It addresses the security risks of running untrusted tool servers locally, the lack of centralized governance for enterprise teams, and the high token costs associated with providing too many tool descriptions to an LLM.

How it works

ToolHive uses a modular architecture consisting of four main parts:

• Gateway: Acts as the entry point, managing authentication (OIDC/OAuth), authorization policies, and tool filtering to reduce token usage.
• Registry Server: A curated catalog of trusted MCP servers that allows teams to discover and deploy verified tools.
• Runtime: Handles the actual execution of MCP servers in isolated containers (via Docker, Podman, or Kubernetes) to ensure security and resource isolation.
• Interfaces: Provides a desktop UI and CLI for users to discover and configure servers, and a Kubernetes Operator for platform engineers to manage them at scale.

Who it’s for

• Developers who want to run MCP servers securely with minimal configuration and lower token costs.
• Platform Engineers who need to deploy and govern MCP infrastructure on Kubernetes.
• Enterprises requiring self-hosted MCP solutions to maintain control over sensitive data and meet compliance requirements.

Highlights

• Container Isolation: Every MCP server runs in its own isolated container to prevent unauthorized local access.
• Token Optimization: Uses semantic tool search and filtering to reduce token usage by up to 85%.
• Enterprise Governance: Includes a Kubernetes Operator for policy declaration, audit logging, and OTel/Prometheus observability.
• Flexible Deployment: Supports local desktop use, cloud-scale Kubernetes clusters, and hybrid setups.