$ buildwith info serena

Serena

#2 in Context & Indexing · Free / OSS

sources differ by section

Serena is an MCP server that gives a coding agent symbol-level retrieval, editing, refactoring, and debugging, closer to an IDE than to line-based search. It does not run the model. Your client (Claude Code, Codex, OpenCode, Gemini CLI, Cursor, a JetBrains assistant, Claude Desktop, or another MCP client) calls Serena's tools. The default backend is free language servers. A paid JetBrains plugin, with a free trial, is the other backend. The README says not to install Serena from an MCP or plugin marketplace, because those commands are outdated. The supported install is uv, then uv tool install -p 3.13 serena-agent, then serena init.

for Developers who already have a coding agent and want it to navigate and edit a large or multi-language repo by symbol. The README's client list covers terminal agents, IDE assistants, and desktop or web MCP clients. You bring the model and the client. Serena is the tool server.

use this if your coding agent needs to find a symbol, see who references it, and edit that symbol, instead of grepping and rewriting whole files.

open oraios.github.io ↗

Features

$ features --list

  • ›Retrieval over language servers: find symbol, a file's symbol overview, find referencing symbols, find declaration, diagnostics. Find implementations is only available for some languages. Search in project dependencies and type hierarchy are marked JetBrains-only
  • ›Symbolic edits on the free backend: replace symbol body, insert before or after a symbol, and safe delete. The README describes these as less error-prone than search-and-replace over file text. Rename of symbols works on language servers. Rename of files and directories, move, and inline are JetBrains-only
  • ›The language-server backend is the default and is described as supporting over 40 languages, including Python, TypeScript, JavaScript, Rust, Go, Java, C#, C and C++, Ruby, PHP, and others listed in the README. Some languages need extra dependencies
  • ›The JetBrains plugin adds dependency indexing, interactive debugging (breakpoints, variables, expressions, execution control), and one IDE process shared by many agent sessions. The docs say it is a backend for the MCP server, not a chat UI inside the IDE
  • ›A memory store for notes that can be shared across sessions, and YAML configuration at global, launch, project, and client-context levels. Inside Claude Code or Codex, the README says the basic file, search, and shell tools are typically disabled because the harness already has them
  • ›v2 adds a REPL-based agent interface. The README points at the docs for how to turn it on

Pricing

$ pricing --tiers

Language servers

Free

uv tool install -p 3.13 serena-agent, then serena init. This is the default backend. The application is GPL-3.0-or-later. SolidLSP, the language-server client library in the same repo, stays MIT. The README says a distribution that combines both, including the serena-agent package, is as a whole under the GPL.

JetBrains plugin

Paid, free trial

Optional backend. The README and the plugin docs call it a paid plugin with a free trial, and say purchases support Serena development. The plugin docs page does not list a price. You still run the Serena MCP server and connect your own client. The plugin is not a chat UI.

Why this rank

$ diff --rank 2

Ranked #2 in Context & Indexing. Serena is the symbol tool: find a symbol, find references, and replace a symbol body through language servers, with no embedding model described in the README. It had 31,273 GitHub stars on 2026-11-11. It ranks second because it does not keep an index of documents or code chunks fresh as sources change, and the free backend stops short of dependency search and several refactors. Sourcegraph ranks third: it searches across an organization's repositories, and its MCP server is an Enterprise plan. The JetBrains plugin that adds those is paid, and the README does not list the price. The scores are unmeasured placeholders and are not shown.

Strengths

$ strengths

  • ›On 2026-11-11 the GitHub API reported 31,273 stars and 2,136 forks. The repository was created 2025-03-23. GitHub's primary language is Python
  • ›Latest GitHub release on that date was v2.1.0, published 2026-10-19. The release notes say 'See change log' and do not list the changes themselves. PyPI serena-agent 2.1.0 was uploaded the same day and requires Python >=3.11 and <3.15
  • ›No embedding model and no vector index. Symbol lookup goes through a language server, or through a JetBrains IDE if you use that plugin, so a search does not add an embedding-API bill
  • ›The README's retrieval table is explicit about what the free backend cannot do, which makes the gap between the two backends checkable

Limitations

$ limits

  • ›The Serena application from v2 onward is GPL-3.0-or-later. The LICENSE file says this is not retroactive: releases up to and including v1.7.0 stay MIT. A combined distribution such as the serena-agent package is under the GPL as a whole
  • ›Find declaration generally does not work for declarations in external dependencies. Find implementations depends on the language server and is missing for some languages. Dependency search, type hierarchy, move, and inline require the paid JetBrains plugin
  • ›The README calls the JetBrains plugin paid, with a free trial, and does not list a price. It also says Rider and CLion are unsupported
  • ›Serena does not index documents, meeting notes, or embeddings, and it does not keep a derived index fresh when files change. It answers the project that is open, through the language server or the IDE
  • ›The README's efficiency claims come from an evaluation prompt the project wrote and then summarized in the agents' own words. The page links a write-up. Those quotes are not an independent measurement of tokens saved

Sources

$ sources --verified

$ buildwith alternatives serena

  • #1 CocoIndexFree / OSS

    use this if you want a coding agent to search a repo by meaning and get back small chunks, or you want an index of code and documents that only reprocesses what changed.

  • #3 SourcegraphPaid

    use this if your team already has Sourcegraph, or will, and you want coding agents to search across repositories and get file ranges back instead of reading the tree themselves.

  • #4 RepomixFree / OSS

    use this if you want one file of a repository, or a compressed signature view of it, to hand to a chat model or an agent, and you do not want to run an index.

sponsor this spot →