lean-ctx

Hybrid Context Optimizer — Shell Hook + MCP Server. Single Rust binary, zero dependencies.

[](https://crates.io/crates/lean-ctx) [](rust/LICENSE)

Website · Install · Quick Start · CLI Reference · MCP Tools · vs RTK

---

lean-ctx reduces LLM token consumption by 89-99% through two complementary strategies in a single binary:

1. Shell Hook — Transparently compresses CLI output before it reaches the LLM. Works without LLM cooperation.
2. MCP Server — 8 tools for cached file reads, dependency maps, entropy analysis, and session metrics. Works with Cursor, GitHub Copilot, Claude Code, Windsurf, and any MCP-compatible editor.

| Operation | Frequency | Standard | lean-ctx | Savings | |---|---|---|---|---| | File reads (cached) | 15x | 30,000 | 195 | -99% | | File reads (map mode) | 10x | 20,000 | 2,000 | -90% | | ls / find | 8x | 6,400 | 1,280 | -80% | | git status/log/diff | 10x | 8,000 | 2,400 | -70% | | grep / rg | 5x | 8,000 | 2,400 | -70% | | cargo/npm build | 5x | 5,000 | 1,000 | -80% | | Test runners | 4x | 10,000 | 1,000 | -90% | | curl (JSON) | 3x | 1,500 | 165 | -89% | | docker ps/build | 3x | 900 | 180 | -80% | | Total | | ~89,800 | ~10,620 | -88% |

> Estimates based on medium-sized TypeScript/Rust projects. MCP cache hits reduce re-reads to ~13 tokens each.

cargo install lean-ctx

git clone https://github.com/yvgude/lean-ctx.git
cd lean-ctx/rust
cargo build --release
cp target/release/lean-ctx ~/.local/bin/

> Add ~/.local/bin to your PATH if needed: > ```bash > echo 'export PATH="$HOME/.local/bin:$PA