Serena

Serena is an open-source MCP (Model Context Protocol) toolkit built by Oraios AI that equips coding agents with IDE-grade semantic understanding of codebases. Licensed under MIT and available on GitHub, it reached 24,000+ stars and version v1.5.1 as of May 2026. Rather than relying on line numbers or primitive text search, Serena operates at the symbol level — enabling agents to navigate, refactor, and edit code the way a seasoned developer using an IDE would.

What It Is

Serena positions itself as "the IDE for your coding agent." It exposes a rich set of tools via the Model Context Protocol, allowing any MCP-compatible LLM client to gain semantic code intelligence without being tied to a specific IDE or editor. The core insight is that agents working with low-level primitives (line numbers, regex search-replace) are slower, less reliable, and more token-hungry than agents equipped with symbol-aware abstractions. Serena bridges that gap by wrapping language server protocol (LSP) backends — or optionally a paid JetBrains plugin — into high-level agent-friendly tools.

How It Works

Serena runs as an MCP server that clients connect to either via a launch command or an HTTP endpoint. It supports a wide range of MCP-compatible clients:

• Terminal-based agents: Claude Code, Codex CLI, OpenCode, Gemini-CLI
• IDE assistant plugins: Copilot, Junie, JetBrains AI Assistant for VSCode, Cursor, and JetBrains IDEs
• Desktop/web clients: Claude Desktop, Codex App, OpenWebUI

Two backend options power the semantic analysis: the default open-source language server (LSP) backend, and the paid Serena JetBrains Plugin that leverages JetBrains IDE analysis engines for deeper refactoring capabilities like move, inline, and propagate-deletions.

Language and Capability Coverage

Using the LSP backend, Serena claims support for over 40 programming languages, including Python, TypeScript, Java, Go, Rust, C/C++, C#, Ruby, Kotlin, Swift, Scala, and many others. Key capability categories include:

• Retrieval: find symbol, symbol overview/file outline, find referencing symbols, find implementations, diagnostics
• Refactoring: rename, move (symbol/file/directory), inline, propagate deletions — with varying availability between backends
• Symbolic editing: replace symbol body, insert before/after symbol, safe delete
• Interactive debugging: exclusive to the JetBrains plugin; supports breakpoints, variable inspection, expression evaluation via a REPL-style interface
• Memory management: a built-in memory system for sharing knowledge across sessions, users, and projects
• Basic utilities: regex search, file read, directory listing, shell command execution (typically disabled when inside an agentic harness that already provides these)

Agent-First Design Philosophy

Serena's design explicitly targets AI agents as its "end users." The README includes an unbiased evaluation methodology where agents perform ~20 routine coding tasks and self-report on the value added by Serena's tools. Vendor-published agent evaluations (Claude Opus 4.6 in Claude Code, GPT 5.4 in Codex CLI, GPT 5.4 in Copilot CLI) independently converge on the conclusion that Serena's semantic tools reduce multi-step, error-prone operations to single atomic calls — particularly for cross-file renames, moves, and reference lookups.

Update: v1.5.1 (May 2026)

The latest release is v1.5.1, published May 18, 2026, with the repository last pushed on May 19, 2026. The project was created in March 2025 and has grown rapidly, accumulating 24,395 GitHub stars and 1,635 forks. The Oraios team explicitly warns against installing Serena from MCP or plugin marketplaces, citing outdated and suboptimal installation commands there — the recommended path is via uv tool install from PyPI. The JetBrains plugin is available on the JetBrains Marketplace (plugin ID 28946). Active development and community contributions have expanded language support significantly, with the open-source community credited for much of the multi-language coverage.