Serena

Text search and whole-file rewrites waste most of what a compiler already knows, so Oraios AI built Serena to hand that knowledge to coding agents. The MIT-licensed toolkit runs as an MCP server backed by real language servers, giving any connected agent semantic retrieval and editing at the symbol level: find a function by name, list every reference across files, rename safely, or insert code relative to a symbol instead of a line number. The LSP backend covers more than 40 languages including Python, Java, JavaScript, Go, Rust, and C++, and a paid JetBrains plugin with a free trial offers an alternative backend that reuses IDE indexing. It plugs into Claude Code, Codex CLI, Gemini CLI, OpenCode, Claude Desktop, Cursor, VSCode, and JetBrains IDEs, and installs with uv tool install serena-agent, with uv as the only prerequisite. The core toolkit runs locally against the codebase with no API dependency of its own and costs nothing. At roughly 28,000 GitHub stars, it has become the standard answer when an agent needs IDE-grade code navigation rather than grep and full-file rewrites.