codebase-memory-mcp vs serena
Code intelligence MCP in pure C: tree-sitter knowledge graph over 158 languages, average repo indexed in milliseconds, sub-ms queries, 10x fewer tokens. Single static binary, zero deps. — versus — The IDE for your coding agent: an MCP toolkit giving symbol-level retrieval, editing, refactoring and debugging over real language servers — or a JetBrains plugin backend.
Same job, opposite engineering bet: codebase-memory-mcp indexes a tree-sitter knowledge graph in pure C for millisecond indexing and sub-millisecond queries; Serena skips the index and asks a language server live, trading startup cost for semantic accuracy.
| codebase-memory-mcp | serena | |
|---|---|---|
| Stars | 40k | 28k |
| Forks | 3.2k | 1.9k |
| Language | C | Python |
| License | MIT | MIT |
| Last activity | 5 days ago | 4 days ago |
| Topics | code-intel, local | code-intel, coding |
| Curated connections | 7 | 4 |
codebase-memory-mcp — the curator's take
The performance ceiling of the code-context-server category: pure C, single static binary, the Linux kernel indexed in 3 minutes, structural queries under a millisecond — with a peer-reviewed preprint (83% answer quality, 10x fewer tokens across 31 repos) instead of vibes. Hybrid LSP adds real type resolution for the 12 languages that matter most, and 43 client surfaces means it plugs into whatever agent you run. NOT semantic search — it answers structural questions (call chains, routes, blast radius), not 'where's the code that does X'; pair it with an embedding tool for that. And note its own disclosure: it writes to your agent config files by design — audit posture is unusually good (SLSA 3, OpenSSF, per-release VirusTotal), use it.
serena — the curator's take
The default answer when an agent is flailing in a large codebase: cross-file renames, moves and reference lookups collapse from a dozen careful text edits into one atomic symbol-level call, because a real language server — not a regex — resolves the code. It bolts onto any MCP client (Claude Code, Codex, Cursor, JetBrains, Gemini CLI, Copilot CLI), and the evaluation methodology is refreshingly honest: they had agents score the tools on ~20 routine tasks and published the prompt. Two things to know: install from their Quick Start, not a plugin marketplace, because the marketplace commands are outdated and the maintainers say so; and the JetBrains backend is a paid plugin, while the free path depends on whichever LSP server your language has.