Mibayy/token-savior
MCP server that gets Claude to 97.9% (188/192) on a real coding benchmark at -80% active tokens and -83% wall time, vs 78.3% plain. Structural code navigation + persistent memory engine. Works with every MCP client. observed · 2026-08-28
Health v2 · maintenance only
79/100
- Activity 97
- Release rhythm 95
- Longevity 11
Flags: young
How is this computed?
round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10) — computed 2026-09-03. Adoption (stars, forks) is never an input.
- gap_med: 0
- age_days: 156
- days_rel: 36
- days_push: 23
- n_releases_24m: 30
Adoption not part of the score
1125 stars · 94 forks observed · 2026-08-28
What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded
Token Savior is an MCP server written in Python that provides AI coding agents like Claude with structural codebase navigation (72 tools for dependency graphs, import cycles, callers, hotspots) plus a persistent memory engine and Bash command rewriting. It claims large token and wall-time savings on coding benchmarks, though its headline benchmark figures are self-reported and not independently verified.
Use cases
- reduce token usage when Claude Code navigates a large codebase
- find callers and dependencies of a function without reading whole files
- detect import cycles in a Python or TypeScript project
- give an AI coding agent persistent memory across sessions
- estimate blast radius of a change before refactoring
- summarize modules and find semantic duplicates in a repo
- speed up AI coding agent tasks with structural code tools
When to choose
- you use Claude Code or another MCP client and want token-efficient codebase navigation
- your agent wastes tokens grepping and reading files to answer structural questions
- you want persistent memory for an AI coding assistant
- you work in Python, TypeScript, Go, Rust, C, or GLSL codebases
When to avoid
- you need independently verified benchmark results before trusting performance claims
- your MCP client defers tool loading so the server's tools never appear to the model
- you need a fully published benchmark harness to reproduce scores
- you don't use MCP-compatible AI coding agents
Facets
service · maturity active
mcp parser developer-tools search-engine caching developer-tools large-language-models parsers programming-languages python cli cross-platform mcp-server code-navigation token-efficiency persistent-memory ai-coding-agents structural-analysis claude code-indexing ai-agents
3 sources
- readme: https://github.com/Mibayy/token-savior · fetched 2026-08-28 · 81981dbd5b05
- homepage: https://mibayy.github.io/token-savior/ · fetched 2026-08-29 · 88ff893763a6
- registry_pypi: https://pypi.org/pypi/token-savior/json · fetched 2026-08-29 · 9999cb09fc84
Member repositories
| Repository | Role | Health v2 |
|---|---|---|
| Mibayy/token-savior | main | 79 |
For agents
markdown · JSON · MCP: product_card(name="Mibayy/token-savior")
Data as of 2026-08-30T08:39:29.467469+00:00 · Report a problem