What is Claude Code Router and should a team use it?

Short answer

Claude Code Router (CCR) is an open source, MIT-licensed project by musistudio, not by Anthropic. It runs a local gateway on your machine (port 3456 by default) that sits between coding agents such as Claude Code, Codex, OpenCode or Grok CLI and the model providers you configure, with routing rules, fallbacks and request logs. Anthropic does not support routing Claude Code to non-Claude models.

What it does

The name suggests a tool for Claude Code alone, but today it is what its README calls “a local model gateway and control plane for coding agents”. It gives each agent one stable local endpoint, and behind that endpoint you manage:

Area What CCR offers, per its README
Agents Profiles for Claude Code, Claude Design, Codex, Grok CLI, Kimi CLI, Kilo Code, OpenCode, Pi, ZCode and WorkBuddy
Providers Presets and custom endpoints: OpenAI, Anthropic, Gemini, OpenRouter, DeepSeek, Moonshot, Mistral, Z.AI and others; single keys or credential pools
Routing Conditions on headers and bodies, rewrites, retries and ordered fallback models
Extras Web search, MCP tools, browser automation and “Fusion” to add vision to models that lack it
Observability Request logs with resolved provider and model, latency, tokens and estimated cost

You start it as a desktop app or with npm install -g @musistudio/claude-code-router and ccr ui. The gateway listens on http://127.0.0.1:3456; you add a provider, start the server, and apply an agent profile.

Why people use it

  1. Cost. Send cheap tasks to a cheaper model and keep the expensive model for hard ones.
  2. Choice. Try open-weight or regional models with the Claude Code interface people already know.
  3. Resilience. Fall back to another provider when one is rate-limited or down.
  4. One place to see usage across several agents.

The common misunderstanding: “it’s still Claude Code”

The interface is Claude Code; the brain may not be Claude. Anthropic’s gateway documentation is explicit that it “doesn’t support routing Claude Code to non-Claude models through any gateway”. Claude Code’s prompts, tools and features are tuned for Claude models, and Anthropic warns that a gateway that doesn’t forward new capabilities “breaks the corresponding features”. So when something works worse through CCR, the cause can be the model, the translation between API formats, or a feature the gateway does not pass on, and nobody upstream will debug that combination for you.

The second misunderstanding is about billing. Anthropic writes that when a gateway credential is active, “the subscription’s usage limits don’t apply” and traffic is “billed per token to whoever owns the credential the gateway forwards”. A router does not turn a subscription into free API capacity.

Model names stay in your hands as well. Claude Code’s configuration docs state that ANTHROPIC_BASE_URL “changes where requests are sent, not which model answers them”. A gateway that is not set up to serve the model Claude Code asks for can reject the request; a router such as CCR avoids that by rewriting the request to a model it does serve. Aliases such as sonnet float to the version recommended for your provider and can resolve to different versions on different providers; to pin one, use the full model name or a variable such as ANTHROPIC_DEFAULT_SONNET_MODEL.

What a team should weigh

Question CCR on each laptop A gateway the company runs
Who holds provider keys? Each developer The gateway; developers get their own revocable credential
Who sees usage and cost? Each developer, locally One place, per person or team
Can you enforce it? No Yes: Claude Code’s managed settings can pin ANTHROPIC_BASE_URL and allow only that endpoint (allowedProviders)
Data path Whatever each person configured Reviewed once

For one developer experimenting, CCR is a reasonable tool. For a team, provider keys on every laptop and prompts flowing to providers nobody reviewed are the real risk. Read is Claude Code safe for company code before rolling anything out, and Claude Code token cost per developer if cost is the reason you are looking at a router.

What this means for a team

Switching models does not by itself reset the session. The next model receives the context the client and gateway forward. Project conventions still live only in whatever files each agent loads, and a router does not create shared memory across separate agents or colleagues. A router moves requests, not knowledge. See shared memory across coding agents.

Where Bombos fits

Claude Code Router decides which model answers; Bombos gives every model and agent the same team memory. It is a plugin underneath Claude Code, Codex, Cursor, Copilot, Grok and your own agents that learns how your team works, draws on what already sits in repos and documents, and shows with every answer what it rests on. It runs in the cloud in Europe or standalone on your own server, and one developer can try it first with their own agents.

Related questions

Is Claude Code Router made by Anthropic?

No. It is an independent open source project on GitHub (musistudio/claude-code-router), MIT-licensed. Anthropic's docs say it does not endorse, maintain or audit third-party gateway products.

How do I install it?

Either the desktop app for macOS, Windows or Linux from the GitHub releases, or the CLI: npm install -g @musistudio/claude-code-router (Node.js 22 or newer), then ccr ui. The management UI opens on port 3458 and the gateway listens on 127.0.0.1:3456.

Can I use Claude Code Router with my Claude subscription?

According to Anthropic's gateway docs, when a gateway credential is active, requests carry that credential instead of your claude.ai login and the subscription limits don't apply; usage is billed per token to whoever owns the credential the gateway forwards.

What is the official alternative for companies?

Anthropic documents LLM gateways for organizations: route Claude Code through a gateway you run (or its own Claude apps gateway), with central credentials, usage tracking, cost controls and audit logging, distributed through managed settings.

Sources

  1. musistudio/claude-code-router (README)
  2. Claude Code Docs: Other LLM gateways
  3. Claude Code Docs: Model configuration
The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.