Codebase Memory MCP vs Serena vs GitNexus vs Claude Context
The short answer
Use codebase-memory-mcp as the default, add Serena for refactoring, and check GitNexus’s licence before using it at work. As of November 15, 2026 (GitHub data read the same day):
| Server | Approach | Licence | Stars | Latest release | Needs external service? | Best for |
|---|---|---|---|---|---|---|
| codebase-memory-mcp (DeusData) | Persistent local knowledge graph of code, docs, config and infrastructure | MIT | 48,651 | v0.12.1 (Oct 16, 2026) | No | Default code memory for any agent |
| Serena (Oraios) | Language-server symbol tools: find, edit, rename, move | GPL-3.0-or-later (SolidLSP part MIT) | 31,389 | v2.1.0 (Oct 19, 2026) | No | Precise refactors in large repos |
| GitNexus | Knowledge graph of symbols, calls and execution flows + web explorer | PolyForm Noncommercial 1.0.0 | 48,551 | v1.6.12 (Sep 12, 2026) | No | Impact analysis on personal / OSS projects |
| Claude Context (Zilliz) | Semantic vector search over the codebase | MIT | 12,694 | npm packages, no GitHub release | Yes: vector DB (Milvus/Zilliz Cloud) + embedding API | ”Find code that does X” in huge monorepos |
All four repositories had commits between November 12 and 15, 2026, so none is abandoned.
What is the best code base memory MCP server? codebase-memory-mcp for most teams: MIT, fully local, no API key, broadest client support.
How they differ
codebase-memory-mcp builds a graph that connects source code to documentation, configuration and infrastructure, so an agent can trace a deployment back to its code or ask what a change could affect. It is pure C, parses 164 languages and formats with vendored grammars, adds “Hybrid LSP” type resolution for supported languages, and its installer sets up 46 client integrations (Claude Code, Codex, Cursor and others). An optional visual companion, CBM Lens, is in open beta (v0.0.1). Our hands-on review covers setup.
Serena gives the agent IDE-style operations at the symbol level — find references, rename across files, move a symbol — through language servers, instead of line numbers and grep. Serena v2 added a REPL-based agent interface. The application is GPL-3.0-or-later; that matters only if you redistribute it, not if you run it locally. See our Serena review.
GitNexus indexes a repository into a graph and exposes query, context, impact, trace, detect_changes and rename tools, plus a browser graph explorer. npx gitnexus analyze also writes AGENTS.md/CLAUDE.md files and Claude Code hooks (use --index-only to skip that). It needs Node.js 22.18+ or 24.11+. The catch is the PolyForm Noncommercial licence. More in our GitNexus review.
Claude Context embeds your code into a vector database and returns semantically related chunks. It is the only one of the four that needs external services: a Milvus or Zilliz Cloud database and an embedding provider key (OpenAI by default), so code chunks leave your machine unless you self-host both. See our Claude Context review.
Picks by situation
| Situation | Pick |
|---|---|
| Company codebase, code must stay local | codebase-memory-mcp |
| Large Java/TypeScript/Python repo, many cross-file refactors | Serena (alongside a graph server) |
| “Which services break if I change this?“ | codebase-memory-mcp, or GitNexus for non-commercial work |
| Natural-language search across millions of lines | Claude Context with self-hosted Milvus |
| Personal side project, want a visual graph | GitNexus |
Graph and symbol servers complement each other: running codebase-memory-mcp for structure and Serena for edits is a common pairing. Keep the number of MCP tools modest, though — every tool description costs context on each turn.
Setup checklist
- Index a branch you care about and ask three questions you already know the answer to (callers of a function, where a config value is used, what a PR touches).
- Compare tokens per task with and without the server on the same prompt.
- Re-index on branch switch or in a git hook; stale graphs mislead agents.
- For onboarding conventions that a graph cannot infer, add an AGENTS.md — see how to onboard AI agents to your codebase and best AI coding assistant for large codebases.
Last verified: November 15, 2026. We describe these tools from their repositories and documentation; GitHub star counts change daily.