ouroboros
Agent OS: the agent gets smarter on its own. We just hold the line: Interview-gated, staged evaluation, budgeted evolution loop. MCP server, 14 runtimes: Claude Code, Codex CLI, Gemini CLI, OpenCode, Copilot, Kiro and more.
- stars
- 5,895
- 30-day movement
- +279/day
- Related entries
- 84
- Connections
- 5
Ouroboros is an MCP server that runs a self-improvement loop for coding agents: an interview-gated, staged evaluation process with a budgeted evolution loop, so the agent gets smarter on its own while you only hold the line. It is a Python toolchain (uv-based) that ships agents.md, .claude/ directory, claude.md, and MCP config, and supports 14 runtimes including Claude Code, Codex CLI, Gemini CLI, OpenCode, Copilot, and Kiro.
Reach for it when you want your coding agent to improve itself through structured, budgeted evolution rather than manual prompt tuning.
Use it to
- Run a budgeted self-evolution loop for a coding agent
- Gate agent changes behind interview and staged evaluation
- Serve the loop to 14 agent runtimes via MCP
- Set up agents.md, claude.md, and MCP config in a repo
- Evaluate LLM agent skill improvements over iterations
For Developers running self-improving coding agents across multiple runtimes
- Role
- mcp-server
- Language
- Python
- Licence
- MIT
- Forks
- 597
- Open issues
- 72
- Last push
- 2026-09-14
- Latest release
- v0.1.0a1 · 2026-01-28
- Skills shipped
- 22