Skip to content
PrismRoute
简体中文

Switch Claude Code and Codex with CC Switch.

CC Switch is a desktop tool that stores several providers for command-line clients such as Claude Code and Codex and switches between them in one click. This page covers both PrismRoute setups: Claude Code with Claude models, and Codex with GPT models.

On this page

Choose a client

The two clients use different request formats and addresses. Pick the client that matches the model family you want, and keep each setup separate.

ClientRecommended modelsAddressRequest format
Claude Codeclaude-sonnet-5, claude-opus-5, claude-haiku-4-5-20251001https://codex.aicpr.icu (no /v1)Anthropic Messages
Codexgpt-5.5https://codex.aicpr.icu/v1OpenAI Responses

Import in one click

Install and open CC Switch, then select a button below. Your browser asks whether to open CC Switch, which then shows an import preview.

Import Claude Code presetImport Codex presetThe links contain no API key
  1. Check the name, address and models in the import preview.
  2. After importing, edit PrismRoute in CC Switch, paste your API key into the API Key field and save.
  3. Select PrismRoute to enable it, then restart Claude Code or Codex in your terminal.

Set up Claude Code

In CC Switch, switch to Claude Code, add a custom provider and fill in the fields below. Enter only the root address; Claude Code appends /v1/messages itself.

FieldValue
Provider namePrismRoute
API KeyYour PrismRoute API key
Addresshttps://codex.aicpr.icu
AuthenticationKeep the default (Bearer / ANTHROPIC_AUTH_TOKEN)
Local routingNot needed
Sonnet modelclaude-sonnet-5
Opus modelclaude-opus-5
Haiku modelclaude-haiku-4-5-20251001
Claude Code address
https://codex.aicpr.icu

Set up Codex

In CC Switch, switch to Codex, add a custom provider, enter your API key and use the content below as config.toml. Codex needs the /v1 address and the Responses request format.

Codex config.toml
model_provider = "prismroute"
model = "gpt-5.5"
model_reasoning_effort = "medium"

[model_providers.prismroute]
name = "PrismRoute"
base_url = "https://codex.aicpr.icu/v1"
wire_api = "responses"

Manual setup without CC Switch

You can also edit each client's own configuration file. Back up the original file before changing it.

For Claude Code, edit ~/.claude/settings.json (%USERPROFILE%\.claude\settings.json on Windows). If the file already has an env block, merge only these entries:

~/.claude/settings.json
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://codex.aicpr.icu",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_PRISMROUTE_API_KEY",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4-5-20251001"
  }
}

For Codex, edit ~/.codex/config.toml and set PRISMROUTE_API_KEY in the terminal where you start Codex:

~/.codex/config.toml
model_provider = "prismroute"
model = "gpt-5.5"
model_reasoning_effort = "medium"

[model_providers.prismroute]
name = "PrismRoute"
base_url = "https://codex.aicpr.icu/v1"
wire_api = "responses"
env_key = "PRISMROUTE_API_KEY"
Terminal environment (macOS / Linux)
export PRISMROUTE_API_KEY="YOUR_PRISMROUTE_API_KEY"

In Windows PowerShell, run setx PRISMROUTE_API_KEY "YOUR_KEY" and open a new terminal.

Known limitations

CaseDetails
GPT models in Claude CodePlain chat works, but the tool list is not passed to GPT, so reading files, editing code and running commands do not work. Claude Code also warns that it does not recognise gpt-5.5 and assumes a 200K context. Use Codex for GPT.
/v1/messages/count_tokensCurrently returns 404. In our tests, chat and tool use still worked normally.
gpt-5.3-codexNo channel is available yet. Use gpt-5.5 in Codex.
Response timeIn our tests, a Codex reply took about 18 to 49 seconds and Claude Code about 8 seconds. Wait for long tasks instead of resubmitting.

For 401, 404 or model permission errors, work through Troubleshooting.

Technical references