LLM Router is an MCP server and hook set that intercepts prompts and routes them to the cheapest model that can handle the task, saving 60-80% on costs. It works with Claude Code, VS Code, Cursor, and many other AI coding tools, automatically offloading simple tasks to cheaper models while reserving premium models for complex work.

llm-router

Active·★ 119·MIT·Updated 2026-10-22

LLM Router is an MCP server plus a set of hooks for AI coding tools. It targets a common waste: using your best model for everything, including simple explanations, file lookups and small edits. Every prompt passes a UserPromptSubmit hook before the top-tier model sees it, and the router sends easy work to cheap or free models and keeps hard work on Claude or another premium model.

The README describes a layered decision: reuse the previous turn's route for replies like "yes" or "go ahead", score the prompt with free heuristics, ask a local Ollama model when the heuristics are unsure, and only then fall back to paid APIs. In subscription mode it follows a free-first chain of Ollama, then Codex, then a paid API. It lists 46 MCP tools and 20+ providers, with budget caps per provider, a dashboard, Prometheus metrics and OpenTelemetry traces. One command installs it for Claude Code, VS Code, Cursor, Codex CLI and other hosts, and Claude Code Pro or Max users can start without any API key.

llm-router

Features

Why choose it

Trade-offs

Maintenance check

Compatibility

Quick start

Use cases

Alternatives

Comments

Most discussed open issues

Deploy on DigitalOcean