Serena icon

Serena

Serena is a free, open-source toolkit that gives AI models like Claude, Gemini, or GPT real power over your codebase — not just suggestions, but actual control.

Instead of guessing or relying on vague search, Serena connects the AI to your code like an IDE would: it can find functions, edit files precisely, run your tests, read logs, and remember what it did.

🧠 What it actually does

  • Reads code structurally using the Language Server Protocol (LSP), not just text search
  • Finds and edits real symbols (e.g. “insert code after validateUser()”)
  • Runs shell commands, like npm test or pytest, with permission
  • Stores memories about your project so it can pick up where it left off
  • Works with free LLMs (Claude, Gemini, Ollama) — no subscription required
  • Runs as an MCP server, meaning tools like Claude Desktop can connect directly to it

💡 For beginners: Why it matters

Most coding assistants (like Copilot or Claude Code) can only "see" the file you're typing in. They don’t really know your whole project. Serena fixes that by acting like a smart bridge between your AI and your entire codebase — kind of like giving your AI its own keyboard, IDE, and terminal.

⚙️ For developers: What makes Serena different

Compared to most AI agents:

  • Reads code: Most use embedding search; Serena uses symbolic LSP (like VSCode)
  • Edits code: Others guess; Serena edits by function or symbol location
  • Shell commands: Others may not support; Serena runs them with permission
  • Memory: Others forget; Serena stores structured project memories
  • Token use: Others burn tokens fast; Serena is efficient and scoped
  • Pricing: Others often require API credits; Serena is 100% free and local

🔌 How to get started (simplified)

  1. Install Python + uv
  2. Run Serena:
    uvx --from git+https://github.com/oraios/serena serena start-mcp-server
    
  3. Connect it to Claude Desktop or Cursor using MCP config
  4. Tell Claude:
    "Activate the project /path/to/my/code"
  5. You're off — the AI can now work inside your codebase.

🎯 Popular use cases

  • Implementing features across multiple files
  • Refactoring large, complex projects
  • Planning changes with “planning” mode
  • Hands-off “vibe coding” with Claude in the loop
  • Using Claude as a coding agent for free

🧪 Fun fact

Claude (even the free tier) supports MCP, so you can use Serena with zero cost and Claude Desktop today. And because Serena runs locally, no code ever leaves your machine.

🔥 What developers are saying

Turns Claude into a real agent — like Cursor or Copilot, but smarter and free. Finally, an LLM that can read, write, and think like a dev — and doesn’t burn tokens. Symbol-level understanding makes Serena ridiculously precise.

Serena Tool Discussions

Sam Moore's avatar
5mo·Hi everyone, I'm a…

Serena review after a week of use

Here is my review of Serena after using it for a week of daily use for regular dev work. The results are coming in slowly.

0
Sam Moore's avatar
6mo·Hi everyone, I'm a…

Claude Code + Serena is the first time I’ve felt like AI can actually do the job

Okay so I’ve actually been using Claude Code + Serena on a real client project (not just playing around with demos), and wow — this is not just another AI tool. This is like, finally, having an actual teammate. Claude is the brain. It’s fast, doesn’t over-explain, and actually gets your codebase. No…

2
claude-code
serena
workflows

Stats on Serena

Pricing and Plans

(Open Source)

Free

Free
  • Free version available

Features

  • MCP Server and agent modes
  • Symbol-level semantic code editing
  • Onboarding and project memory system
  • Shell execution tools with optional safeguards
  • Claude Desktop and IDE integration
  • Support for 10+ languages via LSP
  • Custom contexts and modes
  • Web-based logging dashboard

AI Capabilities

  • Symbolic code analysis and navigation
  • Semantic code editing and transformation
  • Shell command execution
  • Multi-agent orchestration via MCP
  • Custom context and mode switching
  • Project onboarding with memory persistence
  • Cross-IDE support
  • Token-efficient operations

Integrations

Claude Desktop
Claude Code
Cursor
Cline
Roo Code
Windsurf
Agno Agent
Solid-LSP
ChatGPT Desktop (planned)

Platform & System Requirements

Supported Platforms

Linux
macOS
Windows
Developer SDK

System Requirements

Operating System
WINDOWS, MACOS, LINUX
RAM
4GB minimum
Processor
Modern dual-core CPU
Disk Space
200MB + language servers