Use GitHub Copilot as a native Claude endpoint with kRouter

If your company pays for GitHub Copilot, you are already paying for Claude access — Copilot's model picker exposes claude-* variants. The problem is getting that Claude access out of the Copilot UI and into your own tools without losing what makes Claude worth using: tool-use fidelity, thinking blocks, and prompt caching.

kRouter v0.5.114 solves this by routing Claude models on Copilot through Copilot's native Anthropic /v1/messages endpoint instead of the generic /chat/completions path.

Why the endpoint choice matters

Most integrations pipe Copilot Claude through /chat/completions because it is the lowest common denominator. But that path is lossy for Claude:

• It does not surface prompt-cache token counts for Claude models
• It is not the path where cache_control markers actually get injected
• Pivoting Claude requests through OpenAI shape mangles tool-use blocks and thinking blocks

Copilot's /v1/messages endpoint is Anthropic-native. It is the only Copilot endpoint that surfaces prompt-cache counts for Claude and the one that lets caching work. kRouter now routes Claude there.

How kRouter does it

The executor detects Claude models by name (/claude/i) rather than a static registry field — deliberately, because Copilot's live catalog regularly exposes new claude-* variants ahead of any hardcoded list. When it sees a Claude model, it routes to a dedicated messages-endpoint path that:

• Translates the incoming OpenAI-shaped request to Claude format for the shim (chatCore targets openai for GitHub, so requests arrive OpenAI-shaped)
• Forces stream: true upstream, then buffers back to JSON if your client asked for non-streaming
• Strips the internal _toolNameMap field before dispatch (Anthropic returns a 400 on the extra field)
• Sends the required anthropic-version header (a no-op on the other endpoints)
• Translates the Claude SSE response back to OpenAI shape for your client

GPT, Gemini, and Grok models on Copilot are unchanged — they stay on /chat/completions or /responses. Only Claude gets the native-messages treatment.

Wire your Copilot subscription as a Claude provider

npm install -g @sifxprime/krouter
krouter -t

Then in the dashboard:

1. Go to Providers and connect GitHub Copilot via OAuth
2. Point your client at http://localhost:20128
3. Request a Claude model, e.g. gh/claude-sonnet-4.5

Your Claude Code, Cline, or Cursor requests now flow through the Copilot Claude seat you already pay for — with tool-use and thinking blocks intact.

curl http://localhost:20128/v1/chat/completions \
  -H "Authorization: Bearer sk-krouter-local" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gh/claude-sonnet-4.5",
    "messages": [{"role": "user", "content": "Refactor this function for readability"}]
  }'

The cost angle

For a developer whose employer provides Copilot, this is close to free Claude. The seat is a fixed monthly cost you are paying regardless. Routing your agentic Claude traffic through it — instead of burning a personal Anthropic API key — means your out-of-pocket AI spend can drop to zero for the majority of requests.

Stack it into a combo as a tier: Copilot Claude first, a cheap pay-per-token model as overflow, a free tier as the safety net. kRouter's Zenith routing engine picks the healthiest account for each request automatically.

A note on verification

Full honesty: kRouter's test suite covers the routing, headers, config, and — critically — the translator argument order (getting it backwards would silently corrupt every Copilot+Claude response, so it is explicitly tested). But there was no live GitHub Copilot connection on the build instance, so a real /v1/messages round trip was not exercised end to end. The verification is static: build passes, logic is covered, translator signature matches. When you connect a real Copilot account, that is where the live path gets exercised.

Get started

npm install -g @sifxprime/krouter
krouter -t
# Dashboard → Providers → GitHub Copilot → OAuth

See /install for the full setup and /docs/combos for building your fallback stack.