Serena

Overview

Serena gives AI coding agents the same deep code understanding that developers get from their IDE. Instead of searching files with regex, Serena provides semantic code navigation — find all references to a function, explore type hierarchies, jump to definitions, and understand relationships between code components.

How It Works

1. Index your project — Serena builds a semantic model of your codebase
2. Connect via MCP — Any MCP agent can query the index
3. Semantic queries — Find definitions, references, implementations
4. Intelligent editing — Make changes with full impact awareness

Use Cases

• Large codebase navigation — AI understands million-line codebases
• Refactoring — Find all callers before changing a function signature
• Code review — Understand the full impact of a change
• Bug investigation — Trace call chains to find root causes

Getting Started

{
  "mcpServers": {
    "serena": {
      "command": "uvx",
      "args": ["serena", "--project-root", "/path/to/project"]
    }
  }
}

Example

User: "Find all places that call authenticate() and show error handling"

Agent → Serena: find_references("authenticate", include_body: true)
Returns: 8 call sites across 5 files

Agent: "authenticate() is called in 8 places:
- api/users.py:118 — NO error handling ⚠️
- middleware/auth.py:23 — catches all exceptions, logs and re-raises"

Alternatives

• GitHub MCP — Repository-level GitHub operations
• Git MCP — Read-only access to GitHub repos
• Codebase Search — Text-based code search (not semantic)