Spotify Cut Claude Code Tokens 90% For Free
Spotify's Claude Code plugin claims 90% token savings. Here's what that actually measures, why the second model costs you, and the free block that works. Spotify published shunt, a free Claude Code plugin that blocks whole-file reads over 350 lines and routes them through a second model via Portal, claiming 82, 94% savings on bulk reads in its Java monorepo. The 90% measures Claude's token intake in Spotify's own tests using character-count estimates, excluding the worker model's tokens, Portal's cost, and cached-read pricing, so it describes what Claude reads, not your bill. JetBrains' parallel test of rtk, a similar tool, found it reported 96.2 million tokens saved while median session cost rose 7.6% at low effort, showing how self-reported savings diverge from real billing. Shunt's block itself costs nothing and works reliably despite buggy hooks; the paid part is Portal access and the worker model's processing, plus 10, 30 second handoffs. The block can't handle code editing (no line numbers) or safety-critical reasoning. A free alternative keeps the block, pointing Claude toward targeted reads, file search, or local outlines built with tree-sitter, no second model, no Portal, no lag. Aider and code-intelligence plugins give Claude symbol navigation instead of text search. Anthropic's own cost docs recommend both hooks (preprocessing data before Claude reads it) and code intelligence (replacing text search with structure). Nobody has measured the free version's real impact on your bill, so test it against your own work before trusting any tool's headline savings. For builders using Claude Code, anyone optimizing token usage, and anyone skeptical of token-saving benchmarks. Chapters: 0:00 Intro 1:15 Basics 3:18 How it works 5:59 The 90% 9:19 The catch 12:17 Free version 15:12 Silver bullet? Tools & resources mentioned: - shunt (Spotify): https://github.com/spotify/portal-ai-plugins - Portal by Spotify: https://www.spotify.com/business/try-portal/ - Aider: https://aider.chat/ - tree-sitter: https://tree-sitter.github.io/tree-sitter/ - Claude Code: https://claude.ai/ About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #claudecode #tokenoptimization #aitools
