Tutorial for Claude Code + Serena Best practice

[kekkoz92]

Hi, is there any resource or complete tutorial on how to work with Claude Code + Serena? Also about best practices, sub agents, memories vs claude.md, advanced techniques, recent changes?

I've been using Serena for a few months with Claude Desktop and am very happy about it. I want to make the change with Claude Code, but I still don't have a clear picture on how to set up everything correctly.

Or can you give me some pointers and directions so I can research myself but in the right direction?

Thank you for your excellent work on this project!

[MischaPanch]

Hi. The installation of Serena in Claude Code is very simple, see here.

If you want to add some documentation on best practices, we'd be happy to review a PR on that

Some pointers to explore:

1. Each Serena process starts a language server, which can be resource intensive. For working with subagents, it's better to use sse mode and start the Serena server manually, this way you can have one Serena process that multiple agents can use
2. Serena's editing tools are very efficient, but they are not integrated into the IDE like CC's editing tools (that show diffs). You may want to experiment with disabling them, or maybe disabling replace_regex, since this one is the hardest one to read (though it's very powerful). There are different use cases - one where the user wants to review a lot (and see diffs), and one where the agent just runs.
3. You can experiment with adjusting prompts and the active tools using Serena's comprehensive configuration options - by using custom modes and context files, and by overriding the default prompts. See readme on this branch.

Have fun playing around with CC and Serena!

[nickolay-kondratyev]

ℹ️ Note that SSE transport is supported as well, but its use is discouraged. Use Streamable HTTP instead. - current serena README

Is SSE still the suggestion for sub-agents or should Streamable HTTP be used instead for this use case?