Guides
Each guide answers one question end to end, with L2M as the worked example: what the technique is and when you need it, a walkthrough with real settings and commands, how it works, and what it does not do. Everything runs on your own deployment.
| Guide | When you need it |
|---|---|
| Run a self-hosted MCP gateway | Several people or agents call the same MCP servers, and you want one endpoint, per-client tokens and one place to switch a tool off. |
| Govern MCP tools | MCP servers offer tools that change things, and different roles should get different tools. |
| Least privilege from real runs | You want rules that give each agent only the tools it really uses, checked against recorded calls before they apply. |
| Test MCP servers in CI | Servers change their schemas or go down, and you want tests that don't depend on them and a build that fails on breaking changes. |
| CI for AI agents | Prompt, model and tool changes should be regression-tested and gated before they reach production. |
| Human approval for agent tool calls | An agent's write-tool calls should wait for a person, with the decision on record. |
New to L2M? Start with Build your first MCP agent, or look around the live demo first.