oraios/serena: An MCP toolkit giving coding agents semantic retrieval and editing

Serena connects to clients via the Model Context Protocol and uses a Language Server Protocol or JetBrains plugin backend to understand code at the symbol level rather than the line-number level. The language server backend covers more than 40 programming languages and supports finding symbols, references, type hierarchies, and diagnostics; the JetBrains plugin backend additionally offers renaming files and directories, moving, inlining, deleting unused code, and interactive debugging. Version 2 adds a REPL-based agent interaction interface. For developers using terminal tools like Claude Code, Codex CLI, and Copilot CLI, connecting Serena turns cross-file refactoring in large Python, Java, or multilingual monorepos from error-prone multi-step text operations into a single semantic call. The project's own evaluations show multiple models independently reaching the same conclusion across different codebases: symbol-level navigation and cross-file refactoring are the biggest gains, while small text edits can still use the client's built-in capabilities. Installation is via uv to install serena-agent; note that the official guidance explicitly warns against installing from MCP or plugin marketplaces, as their install commands are outdated. Serena also offers composable multi-layer YAML configuration and a memory management system for long-running agent workflows, sharing knowledge across sessions, users, and projects.