ToolHive
Runs each MCP server in its own container with a proxy, registry, and secrets management.
About
Running an MCP server straight from npx hands it the host shell; Stacklok's ToolHive instead launches each server in its own container behind an HTTP or SSE proxy, so tools get isolation, network controls, and encrypted secret management by default. The Apache-2.0 project, written in Go, ships three ways: a CLI and a desktop app for individual developers, and a Kubernetes operator for platform teams running MCP infrastructure fleet-wide. A built-in registry curates trusted servers, OIDC and OAuth handle single sign-on, and OpenTelemetry plus Prometheus provide monitoring; the proxy layer also claims token usage reductions of up to 85 percent through optimization. It works with every major MCP client, from Claude Code and Cursor to custom agents, and requires only a container runtime such as Docker or Podman on the local machine. With around 2,000 GitHub stars and more than 4,000 commits, it has become one of the main answers to the emerging question of how to run third-party MCP servers without trusting them completely.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- AI Agents & Orchestration
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Easy (2/5)
- License
- Apache-2.0
- Added
- Aug 24, 2026
Related Tools
Multi-agent framework for building and studying societies of communicating LLM agents at scale.
Super-agent harness orchestrating sub-agents, memory, and sandboxes for research and automation.
TypeScript agentic operating system with plugins, memory, and connectors for major messaging platforms.
Code-first, model-agnostic toolkit for building, evaluating, and deploying AI agents.
Autonomous agent that plans queries, searches the web and local files, and writes cited reports.
Platform for building and running autonomous agents that execute complete workflows from goals.