Observal

Why we included this project

Once a team starts building its own MCP servers and skills, along with custom agent configurations, those pieces tend to end up scattered across private repos and chat threads, with nobody able to tell which ones are actually being used. Observal gives you a self-hosted place to register and share them, bundling MCP servers, skills, hooks, prompts, and sandboxes into a versioned agent package, then writes the configuration for whichever coding harness your developers run, whether that's Claude Code, Cursor, Codex, or another one. Beyond the registry, it captures session and adoption data so you can see which agents and tools are helping and which are quietly causing problems. Session traces also come in handy for debugging and audits. That feedback loop is what most teams are missing once agent usage spreads, and if you are responsible for governing internal AI tooling, Observal is worth evaluating as a control plane rather than another agent framework.