Claude Code Router with Plan2API

Settings at a glance

These are the values you'll enter in CCR's Add Provider dialog and your Claude Code profile.

Set up CCR with Plan2API

1. Sign in with a Google account at plan2api.com/connect, connect a Claude Pro or Max, ChatGPT, Google AI Pro or Ultra, or Kimi Code subscription, and copy your p2a_ API key.
2. Install the Claude Code Router CLI with npm install -g @musistudio/claude-code-router on Node.js 22 or later.
3. Run ccr ui to start the background service on http://127.0.0.1:3456 and open the management UI in your browser at http://127.0.0.1:3458 (or use ccr ui --no-open on a headless machine).
4. Open Providers, click Add Provider, pick Other / custom API endpoint, enter Plan2API as the Name, https://api.plan2api.com/v1 as the API endpoint and your p2a_ key, select the models you want to expose, and run Check Connection on one model since each test sends a real request that counts against your request limits.
5. Go to Agent Config, add a Claude Code profile, set its Config name (for example Claude Code Work), and pick Plan2API and claude/claude-opus-5-5 as its Model.
6. Launch Claude Code through the profile with ccr "Claude Code Work". If you use the CCR Desktop app instead of the npm CLI, the terminal button on the profile card copies ccr-app "Claude Code Work" for the CLI, and the play button opens Claude App.
7. Run /model inside Claude Code to verify the available models and check CCR's Request logs to see your requests come through.

# Install Claude Code Router (Node.js 22+)
npm install -g @musistudio/claude-code-router

# Open the management UI at http://127.0.0.1:3458
ccr ui

# Start Claude Code through your CCR profile
ccr "Claude Code Work"

Map Claude Code's model tiers

Your profile's Model setting is the default, while the optional Fable, Opus, Sonnet and Haiku overrides decide which Plan2API model actually answers when Claude Code asks for one of those tiers. Putting GPT, Gemini or Kimi models on some of those tiers spreads the work across your connected Claude and ChatGPT subscriptions instead of burning through one plan's quota. Every model ID you assign to a tier also needs to be checked in your Plan2API provider's Models list in CCR, or CCR won't route to it.

What works and what doesn't

• Streaming, tool and function calling, and reasoning_effort (or reasoning.effort) pass through Plan2API on all four providers.
• Image input works on Claude, Gemini and Kimi models, plus Codex models that accept images, with up to 32 image URLs per request at up to 20 MB each.
• CCR has to connect to Plan2API using the OpenAI Chat Completions protocol at POST /v1/chat/completions, because Plan2API doesn't serve the Anthropic /v1/messages or OpenAI /v1/responses endpoints.
• JSON mode (response_format set to json_object or json_schema) works on Codex, Kimi and Gemini models, but returns HTTP 400 unsupported_parameter on Claude models.
• Anthropic doesn't support running Claude Code against non-Claude models through a gateway, so routing to other models is an unofficial setup.
• Every tool-call round trip counts as a separate request against your Claude usage limits and your Plan2API plan, so the Free plan's 50 chat requests a day (resetting at 00:00 UTC) is fine for a quick test, not a full workday.

Troubleshooting

The table below covers the most common errors from CCR and Plan2API and how to fix them.

curl https://api.plan2api.com/v1/models \
  -H "Authorization: Bearer p2a_your_key"

Where CCR v3 keeps its config

If you're looking for a claude code router config file to edit by hand, CCR v3 doesn't use config.json anymore. Providers, agent profiles and routing rules live in a SQLite database at ~/.claude-code-router/config.sqlite (%APPDATA%\claude-code-router\config.sqlite on Windows, /data/.claude-code-router/config.sqlite in Docker), and CCR reads a legacy config.json only once to migrate old settings.

Older guides that tell you to edit Providers, api_base_url and transformers blocks in JSON are describing the pre-v3 format. In v3, you put the endpoint, key and model list into the Providers page in the UI at http://127.0.0.1:3458, and configure first-match routing rules, fallbacks and per-key credential pool limits like {"rpm": 60, "tpm": 100000} right in the browser.