Integrations with Agents

This page explains how to connect common agents and developer tools to the Site URL, including API BaseURL, API keys, model IDs, and practical client configuration.

Unified rules:

API BaseURL: use the value from the “API BaseURL” section on the Overview page, for example https://cn-api.tokenfor.me for the CN2 region
OpenAI / xAI-compatible endpoints: usually use <API BaseURL>/v1
Anthropic / Claude Code: usually use <API BaseURL> and let the client append /v1/messages
Gemini-compatible endpoints: usually use <API BaseURL>/v1beta
API key / token: the key created in the Site URL console
One key is bound to one vendor group. Use separate keys for GPT, Claude, Gemini, and Grok groups when they are configured separately

Default models used in this page:

GPT: gpt-5.6-sol
Claude: claude-opus-5
Gemini: gemini-3-pro-preview

Generic HTTP template

If a tool supports OpenAI-compatible APIs, first verify the route with a simple request:

http

POST <API BaseURL>/v1/chat/completions
Authorization: Bearer <Your_API_Key>
Content-Type: application/json

{
  "model": "gpt-5.6-sol",
  "messages": [
    { "role": "user", "content": "Hello, test whether the Site URL is working." }
  ]
}

For the CN2 region, the full request URL is:

text

https://cn-api.tokenfor.me/v1/chat/completions

Codex (ChatGPT) Configuration

Codex reads model provider settings from ~/.codex/config.toml. For TokenFor, use the OpenAI Responses wire API.

macOS

Edit the config file:

bash

nano ~/.codex/config.toml

Add or update:

toml

model_provider = "tokenforme"
model = "gpt-5.6-sol"
model_reasoning_effort = "high"
disable_response_storage = true

[model_providers.tokenforme]
name = "tokenfor.me"
base_url = "https://cn-api.tokenfor.me/v1"
wire_api = "responses"
requires_openai_auth = false
env_key = "Tokenforme_KEY"

Then put the key in your shell environment:

bash

nano ~/.zshrc

bash

export Tokenforme_KEY="sk-your-key"

Reload the shell config and start Codex:

bash

source ~/.zshrc
codex

If you are not using the CN2 region, replace https://cn-api.tokenfor.me in base_url with the API BaseURL from the Overview page.

Windows

Edit:

text

%USERPROFILE%\.codex\config.toml

Use the same TOML config:

toml

model_provider = "tokenforme"
model = "gpt-5.6-sol"
model_reasoning_effort = "high"
disable_response_storage = true

[model_providers.tokenforme]
name = "tokenfor.me"
base_url = "https://cn-api.tokenfor.me/v1"
wire_api = "responses"
requires_openai_auth = false
env_key = "Tokenforme_KEY"

Set the user environment variable in PowerShell:

powershell

[Environment]::SetEnvironmentVariable("Tokenforme_KEY", "sk-your-key", "User")
$env:Tokenforme_KEY = "sk-your-key"
codex

After changing persistent environment variables, reopen the terminal so Codex can read them.

Claude Code Configuration

Claude Code uses an Anthropic Messages-style endpoint. Use the API BaseURL without /v1, for example https://cn-api.tokenfor.me, so the client does not accidentally call /v1/v1/messages.

macOS

Option 1: use shell environment variables.

bash

nano ~/.zshrc

bash

export ANTHROPIC_BASE_URL="https://cn-api.tokenfor.me"
export ANTHROPIC_AUTH_TOKEN="sk-your-key"
export ANTHROPIC_MODEL="claude-opus-5"

bash

source ~/.zshrc
claude

Option 2: use Claude Code settings.

bash

mkdir -p ~/.claude
nano ~/.claude/settings.json

json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://cn-api.tokenfor.me",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-key",
    "ANTHROPIC_MODEL": "claude-opus-5"
  }
}

Windows

Set user environment variables in PowerShell:

powershell

[Environment]::SetEnvironmentVariable("ANTHROPIC_BASE_URL", "https://cn-api.tokenfor.me", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_AUTH_TOKEN", "sk-your-key", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_MODEL", "claude-opus-5", "User")

$env:ANTHROPIC_BASE_URL = "https://cn-api.tokenfor.me"
$env:ANTHROPIC_AUTH_TOKEN = "sk-your-key"
$env:ANTHROPIC_MODEL = "claude-opus-5"
claude

You can also edit:

text

%USERPROFILE%\.claude\settings.json

json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://cn-api.tokenfor.me",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-key",
    "ANTHROPIC_MODEL": "claude-opus-5"
  }
}

If your Claude Code version only reads ANTHROPIC_API_KEY, put the same key there instead of ANTHROPIC_AUTH_TOKEN. Avoid mixing different Claude keys on the same machine.

CC Switch Configuration

CC Switch is useful when you want one place to manage providers for Claude Code, Codex, Gemini CLI, OpenClaw, Hermes, and similar tools. Get one target tool working first, then move the route into CC Switch.

To add a TokenFor provider in CC Switch:

Open CC Switch and click +
Prefer App-specific provider
Choose a custom provider for the current target app
Enter the API key and endpoint
Fetch models or enter the model ID manually
Save and switch to the provider
Reopen the terminal used by the target tool

Common values:

Target tool	Endpoint	Model	API format
Codex	`https://cn-api.tokenfor.me/v1`	`gpt-5.6-sol`	OpenAI Responses
Claude Code	`https://cn-api.tokenfor.me`	`claude-opus-5`	Anthropic Messages
Gemini	`https://cn-api.tokenfor.me/v1beta`	`gemini-3-pro-preview`	Gemini

For Codex custom format, confirm that the generated ~/.codex/config.toml contains:

toml

model_provider = "tokenforme"
model = "gpt-5.6-sol"

[model_providers.tokenforme]
base_url = "https://cn-api.tokenfor.me/v1"
wire_api = "responses"
requires_openai_auth = false
env_key = "Tokenforme_KEY"

For Claude custom format, confirm that the generated environment block contains:

json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://cn-api.tokenfor.me",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-key",
    "ANTHROPIC_MODEL": "claude-opus-5"
  }
}

Hermes Configuration

Hermes stores its main config in ~/.hermes/config.yaml. Put secrets in ~/.hermes/.env. For a first working GPT route, configure it as an OpenAI-compatible Responses endpoint.

Edit the secret file:

bash

nano ~/.hermes/.env

bash

OPENAI_API_KEY=sk-your-key

Edit the model config:

bash

nano ~/.hermes/config.yaml

yaml

model:
  provider: openai
  default: gpt-5.6-sol
  base_url: "https://cn-api.tokenfor.me/v1"
  api_mode: responses

You can also switch models inside a Hermes session:

text

/model gpt-5.6-sol

For Claude or Gemini models in Hermes, choose the matching provider type in Hermes model settings and use:

Claude: claude-opus-5
Gemini: gemini-3-pro-preview

Config changes usually apply to new sessions. Restart Hermes or switch models inside the session if an existing chat keeps using the old model.

Cherry Studio Configuration

Cherry Studio supports custom model providers. The usual path is:

text

Settings -> Model Services -> Add

OpenAI-compatible / GPT

Add a custom provider:

Field	Value
Provider type	OpenAI or OpenAI Compatible
API address	`https://cn-api.tokenfor.me`
API Key	`sk-your-key`
Model	`gpt-5.6-sol`

Some Cherry Studio versions append /v1/chat/completions automatically. If the full URL or an extra /v1 fails, check whether the actual request became /v1/v1/chat/completions.

Claude

Add an Anthropic provider:

Field	Value
Provider type	Anthropic
API address	`https://cn-api.tokenfor.me`
API Key	`sk-your-key`
Model	`claude-opus-5`

Gemini

Add a Gemini provider:

Field	Value
Provider type	Gemini
API address	`https://cn-api.tokenfor.me/v1beta`
API Key	`sk-your-key`
Model	`gemini-3-pro-preview`

After saving, run Check/Test Connection. If models are not fetched automatically, add the model IDs manually.

OpenClaw Configuration

OpenClaw can use separate providers for GPT, Claude, Gemini, and Grok. One key is bound to one vendor group, so use the matching key for each group.

GPT

json

{
  "models": {
    "providers": {
      "tokenforme-gpt": {
        "baseUrl": "<API BaseURL>/v1",
        "apiKey": "sk-your-key",
        "models": [
          {
            "id": "gpt-5.6-sol",
            "name": "gpt-5.6-sol",
            "api": "openai-responses",
            "reasoning": true,
            "input": ["text"]
          }
        ]
      }
    }
  }
}

Claude

json

{
  "models": {
    "providers": {
      "tokenforme-claude": {
        "baseUrl": "<API BaseURL>/v1",
        "apiKey": "sk-your-key",
        "models": [
          {
            "id": "claude-opus-5",
            "name": "claude-opus-5",
            "api": "anthropic-messages",
            "reasoning": true,
            "input": ["text"]
          }
        ]
      }
    }
  }
}

Gemini

json

{
  "models": {
    "providers": {
      "tokenforme-gemini": {
        "baseUrl": "<API BaseURL>/v1beta",
        "apiKey": "sk-your-key",
        "api": "google-generative-ai",
        "models": [
          {
            "id": "gemini-3-pro-preview",
            "name": "gemini-3-pro-preview",
            "api": "google-generative-ai",
            "reasoning": false,
            "input": ["text", "image"]
          }
        ],
        "authHeader": true,
        "request": {
          "headers": {
            "Authorization": "Bearer ${models.providers.tokenforme-gemini.apiKey}"
          }
        }
      }
    }
  }
}

Grok

json

{
  "models": {
    "providers": {
      "tokenforme-grok": {
        "baseUrl": "<API BaseURL>/v1",
        "apiKey": "sk-your-key",
        "models": [
          {
            "id": "grok-4.5",
            "name": "grok-4.5",
            "api": "openai",
            "reasoning": true,
            "input": ["text", "image"]
          }
        ]
      }
    }
  }
}

Continue Plugin

Example Continue config for VS Code:

yaml

name: TokenFor Models
version: 0.0.1
schema: v1

models:
  - name: gpt-5.6-sol
    provider: openai
    model: gpt-5.6-sol
    apiKey: "sk-your-key"
    apiBase: "<API BaseURL>/v1"

  - name: claude-opus-5
    provider: anthropic
    model: claude-opus-5
    apiKey: "sk-your-key"
    apiBase: "<API BaseURL>"

  - name: gemini-3-pro-preview
    provider: gemini
    model: gemini-3-pro-preview
    apiKey: "sk-your-key"
    apiBase: "<API BaseURL>/v1beta"

If your Continue version does not support one of these provider types, switch that model to a provider type supported by your local version and verify it with a short test request.

Other Tools

For Qoder, Cursor, and other tools that support custom model APIs, the usual flow is:

Find the Model, Provider, API, or Endpoint settings
Use <API BaseURL>/v1 for OpenAI-compatible tools
Use <API BaseURL> for Anthropic-compatible tools
Use <API BaseURL>/v1beta for Gemini-compatible tools
Paste the API key created in the Site URL console
Select or enter a model ID, such as gpt-5.6-sol, claude-opus-5, or gemini-3-pro-preview
Save and send a short test message

If the tool has Test Connection, Check, or Fetch Models, use it before running long tasks.

Debugging & Troubleshooting

Authentication failure: check whether the key is correct, enabled, and belongs to the vendor group for the selected model
Connection failure: make sure the BaseURL is the API BaseURL from the Overview page, not a website URL
Path failure: check whether /v1 was appended twice, especially in Cherry Studio and Claude Code
Model not found: confirm the model is enabled in the console and use the exact model ID
CC Switch change not applied: reopen the target CLI terminal and confirm it reads the same config directory
Codex still uses official OpenAI: confirm requires_openai_auth = false, env_key, and the environment variable name match exactly

Integrations with Agents ​

Generic HTTP template ​

Codex (ChatGPT) Configuration ​

macOS ​

Windows ​

Claude Code Configuration ​

macOS ​

Windows ​

CC Switch Configuration ​

Hermes Configuration ​

Cherry Studio Configuration ​

OpenAI-compatible / GPT ​

Claude ​

Gemini ​

OpenClaw Configuration ​

GPT ​

Claude ​

Gemini ​

Grok ​

Continue Plugin ​

Other Tools ​

Debugging & Troubleshooting ​

Integrations with Agents

Generic HTTP template

Codex (ChatGPT) Configuration

macOS

Windows

Claude Code Configuration

macOS

Windows

CC Switch Configuration

Hermes Configuration

Cherry Studio Configuration

OpenAI-compatible / GPT

Claude

Gemini

OpenClaw Configuration

GPT

Claude

Gemini

Grok

Continue Plugin

Other Tools

Debugging & Troubleshooting