Serena

What oraios/serena MCP server does

The oraios/serena MCP server extends an MCP-compatible coding client with semantic programming tools. Instead of treating source files as plain text, it works with symbols and their relationships, allowing an agent to inspect structure, locate definitions and references, modify code, and perform refactoring-oriented tasks.

The project is intended for workflows where an AI agent needs IDE-like code understanding. It can help an agent navigate large or complex repositories without first reading every file and provides diagnostics or inspections through the selected analysis backend. Serena supplies the tools; the connected language model remains responsible for deciding which tools to call and how to combine their results.

How it works

Serena uses one of two analysis backends. The default option integrates language servers that implement the Language Server Protocol. The README lists support for more than 40 languages, including Python, Java, JavaScript, TypeScript, Go, Rust, C#, C/C++, Kotlin, PHP, Ruby, Scala, Swift, Terraform, YAML, and others.

The alternative is the Serena JetBrains Plugin. This backend uses the code analysis and editing capabilities of a JetBrains IDE and supports languages and frameworks handled by those IDEs. IntelliJ IDEA, PyCharm, Android Studio, WebStorm, PhpStorm, RubyMine, and GoLand are listed as supported examples; Rider and CLion are identified as unsupported.

MCP provides the connection between Serena and the client. A client can start the server using a launch command, or the server can be started independently in HTTP mode and exposed through its URL.

Tools and capabilities

The documented capabilities include:

• Finding symbols and showing a file's symbol outline
• Finding symbols that reference another symbol
• Locating declarations and implementations
• Querying external projects
• Searching project dependencies when using the JetBrains backend
• Inspecting type hierarchies through the JetBrains backend
• Returning diagnostics and inspections
• Editing and refactoring code using semantic context
• Supporting a memory system for longer-running agent workflows

Backend coverage differs. Some operations depend on what a particular language server supports, and declarations in external dependencies generally cannot be found through the language-server backend. The JetBrains backend provides additional dependency and type-hierarchy capabilities.

Setup and configuration

Configure the oraios/serena MCP server in an MCP client by supplying the server's launch command, or run it in HTTP mode and provide the resulting endpoint URL. The repository directs users to its Quick Start instructions for the current setup procedure rather than relying on marketplace installation commands.

Serena supports a multi-layered configuration system so its behavior can be adapted to different projects and workflows. The supplied material does not specify required environment variables, a universal launch command, or client-specific configuration snippets.

Limitations and notes

The language-server backend is free and open source, while the Serena JetBrains Plugin is paid and includes a free trial. Language support and semantic operations vary by backend and by the capabilities of the underlying language server. The README also notes that an LLM is required to orchestrate the actual coding work; Serena does not independently replace the client or model.

MCP support is described for terminal clients, IDEs and plugins, and desktop or web clients. Examples include Claude Code, Codex, OpenCode, Gemini-CLI, VS Code integrations, Cursor, JetBrains integrations, Claude Desktop, Codex App, and OpenWebUI.