Graft vs tokensave
Context layer for large codebases: a graph of plain-English markdown nodes — no embeddings, no index — agents read like any repo file. Claude Code hooks + MCP; 42% fewer tokens in its bench. — versus — Code-intelligence MCP server for coding agents — a pre-indexed semantic graph (libSQL + FTS5) they query instead of grepping: symbols, callers, impact radius in one call. 100% local, 50+ languages.
Both replace blind grepping with a pre-built codebase understanding: tokensave is a queryable semantic index (libSQL + FTS5) behind MCP tools; Graft is prose-first — explanation files injected via hooks, no query API to learn.
| Graft | tokensave | |
|---|---|---|
| Stars | 437 | 508 |
| Forks | 35 | 49 |
| Language | TypeScript | Rust |
| License | MIT | MIT |
| Last activity | today | today |
| Topics | coding, rag | coding, local |
| Curated connections | 2 | 11 |
Graft — the curator's take
The anti-embeddings bet is the story: the codebase map is prose files — senior-engineer explanations, linked and greppable — so agents consume it with the file tools they already have, and a stale map shows up as a diff in review instead of rotting in a vector store. The 162-run controlled benchmark holding correctness equal is more rigor than this category usually shows; still self-run. Claude Code gets the deep integration (hooks, statusline, background rebuild), everyone else gets MCP. When NOT: nodes are LLM-written prose — building them on a big repo costs real tokens, and prose can drift subtly wrong in ways a symbol graph can't; spot-check the nodes on your hot paths. ★437 and young, with NanoNets behind it.
tokensave — the curator's take
Use it to stop agents burning tokens on grep/glob/read exploration: one MCP call returns the symbols, relationships and snippets a task needs, and it's local (libSQL, no cloud). Broad reach — 50+ languages, 12+ agent integrations. NOT worth it on small repos where a couple of greps suffice, and it's an index you must keep fresh (re-index on change) — a stale graph misleads the agent. 80+ tools is a lot of surface; most tasks touch a handful.