# Portal AI Plugins (Spotify shunt)

> Updated 2026-09-11 · type: tool · category: mcp-agents · status: active · rev 1

Spotify's shunt plugin routes an agent's bulk reads and boilerplate writes to cheaper models via hooks, cutting Claude Code token usage ~90% in its benchmarks.

- Open source: yes (Apache-2.0)
- Self-hostable: yes
- Pricing model: free
- Best for: Teams running coding agents at scale who want to stop paying frontier-model rates for file I/O and boilerplate — the plugin pattern ports to any agent stack with hooks.
- Last verified: 2026-09-11

- **Canonical:** https://gtmstacker.com/registry/tool/spotify-shunt/
- **Source:** [github · spotify/portal-ai-plugins](https://github.com/spotify/portal-ai-plugins)
- **Tags:** mcp-agents, agent-ops, claude-code, token-efficiency, cost-optimization, plugins
- **Repository:** https://github.com/spotify/portal-ai-plugins

## Is Portal AI Plugins (Spotify shunt) open source?

Yes, Portal AI Plugins (Spotify shunt) is open source under the Apache-2.0 license.

## How much does Portal AI Plugins (Spotify shunt) cost?

Portal AI Plugins (Spotify shunt) is free to use.

## Can I self-host Portal AI Plugins (Spotify shunt)?

Yes, Portal AI Plugins (Spotify shunt) can be self-hosted (the source is available under the Apache-2.0 license).

## Alternatives & related

- [TeamAI CLI](https://gtmstacker.com/registry/tool/teamai-cli/)


---

Open-source Claude Code plugin set (Apache-2.0) from Spotify's Portal platform: the 'shunt' plugin routes I/O-heavy agent work — bulk file reads and boilerplate generation — to cheaper worker models via PreToolUse hooks, cutting Claude token usage by ~90% in Spotify's benchmarks on a Java monorepo.

## Provenance

- Apache-2.0 independently WebFetch-verified 2026-09-11 (1.2k★, official Spotify org). The ~90% token-reduction figure is Spotify's own benchmark, published on Spotify Engineering ("Portal by Spotify cut my Claude Code token usage by 90%", Sep 2026) — vendor-reported, one Java monorepo.
- Surfaced twice in the 2026-09-11 pass: the viral-posts brief (macro-tier X post) and press coverage. Curated from the GTM Stacker signal registry (2026-09-11 pass: daily pull + viral-posts brief); license independently WebFetch-verified 2026-09-11.

## Why it matters for a GTM stack

Most coding-agent spend is not reasoning, it is I/O. Routing bulk reads and predictable codegen to a cheap worker model is the same build-vs-buy math every lean team runs on SaaS seats — applied to model tokens. The hook pattern (intercept expensive tool calls, delegate to a cheaper executor) is portable to any agent doing GTM ops work, not just coding.