Claude Code

For a fresh machine, follow the Claude Code integration guide for native installation, proxy troubleshooting and PATH setup. The native installer needs no Node.js. The npx skills command below is optional and requires Node.js and an existing agent.

TIP

Let the skill configure it

With the ModelFold skills installed, you can simply ask your agent "Configure Claude Code to use ModelFold" and it follows the steps on this page. You can also configure it by hand as described below.

npx skills add https://modelfold.ai --skill modelfold-chat-agents -g -a claude-code -a codex -a grok -y

Key rules

1. Discover models with GET https://modelfold.ai/anthropic/v1/models, not from memory or training data. This endpoint gives the list Claude Code can actually run.
2. Authenticate with Authorization: Bearer. Use this recommended header for requests, including the model listing. A header literally named apikey is rejected with 401. Claude Code reads this header from ANTHROPIC_AUTH_TOKEN and adds the Bearer prefix itself, so put the raw key in the variable. ANTHROPIC_API_KEY is sent as X-Api-Key; ModelFold accepts the raw key as well. Prefer setting only ANTHROPIC_AUTH_TOKEN.
3. Set ANTHROPIC_BASE_URL to https://modelfold.ai/anthropic. Claude Code appends /v1/messages itself. Listing and requests share this prefix: GET /anthropic/v1/models and POST /anthropic/v1/messages.

List available models

export MODELFOLD_API_KEY="your_MODELFOLD_API_Key"
curl -s -H "Authorization: Bearer $MODELFOLD_API_KEY" \
  https://modelfold.ai/anthropic/v1/models |
  jq -r '.data[] | [.id, .display_name, (.max_input_tokens | tostring),
                    (.max_tokens | tostring)] | @tsv'

On Windows (PowerShell), send the same request with curl.exe:

curl.exe -s -H "Authorization: Bearer $env:MODELFOLD_API_KEY" `
  https://modelfold.ai/anthropic/v1/models | jq -r '.data[].id'

The response has the Anthropic-compatible shape. .data[] is the list: id is the model name Claude Code sends, and display_name, max_input_tokens and max_tokens describe what the model accepts. type and created_at are boilerplate. If has_more is true, set after_id to last_id and request the next page.

Use ids returned by this endpoint when picking a specific model. Re-run the listing every time instead of reusing an earlier answer; the model set changes.

Configure

Claude Code takes the base URL, credential and model from environment variables:

export ANTHROPIC_BASE_URL=https://modelfold.ai/anthropic
export ANTHROPIC_AUTH_TOKEN=$MODELFOLD_API_KEY

ANTHROPIC_BASE_URL is the root that requests are sent to. Claude Code appends /v1/messages itself, so the value is https://modelfold.ai/anthropic, not the full request URL https://modelfold.ai/anthropic/v1/messages.

ANTHROPIC_AUTH_TOKEN holds the key. Claude Code sends it as Authorization: Bearer. Do not add Bearer to the value, or the header becomes Bearer Bearer ….

The other credential variable is ANTHROPIC_API_KEY, sent as X-Api-Key. ModelFold accepts the raw key there as well.

Set ANTHROPIC_MODEL to an id returned by the listing, or use --model <id>. Do not assume built-in aliases resolve to an available platform model.

The same variables can go into the env block of a settings file:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://modelfold.ai/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "…"
  }
}

When a variable is set both in the shell and in a settings file, the settings file wins. Do not put credentials in the project-level .claude/settings.json, because that file is committed to the repository. Use the user-level file or .claude/settings.local.json.

macOS and Linux

The user-level file is ~/.claude/settings.json. Export the variables in the same shell that starts claude, or add them to your shell profile:

export ANTHROPIC_BASE_URL=https://modelfold.ai/anthropic
export ANTHROPIC_AUTH_TOKEN=$MODELFOLD_API_KEY

Windows (PowerShell)

The file is %USERPROFILE%\.claude\settings.json, with identical content.

$env:ANTHROPIC_BASE_URL = "https://modelfold.ai/anthropic"
$env:ANTHROPIC_AUTH_TOKEN = $env:MODELFOLD_API_KEY
setx ANTHROPIC_BASE_URL "https://modelfold.ai/anthropic"
setx ANTHROPIC_AUTH_TOKEN "$env:MODELFOLD_API_KEY"

setx applies to windows opened afterwards. Set $env: as well to use it right away.

Verify the setup

Start claude in a shell where these variables are set and run /status. The status page should show Anthropic base URL as https://modelfold.ai/anthropic, and an auth line naming ANTHROPIC_AUTH_TOKEN (or ANTHROPIC_API_KEY if that is what you used). Then send a message.

If an Anthropic account login appears, the variables did not reach this process. Fully quit the terminal and reopen it from a shell where the variables are exported.

claude --debug prints the requests Claude Code actually sends.

Common pitfalls

1. A header named apikey. ModelFold returns 401. Prefer putting the raw key in ANTHROPIC_AUTH_TOKEN. ANTHROPIC_API_KEY also accepts a raw key.
2. The full request URL in ANTHROPIC_BASE_URL. Claude Code appends /v1/messages itself, so a value ending in /v1 or /v1/messages becomes /v1/v1/messages.
3. model not found. Claude Code asked for an id this endpoint does not have. Run /model and pick one from the list.
4. Credentials in the project-level .claude/settings.json. That file is committed. Use the user-level file or .claude/settings.local.json.
5. Bearer inside ANTHROPIC_AUTH_TOKEN. Claude Code adds the prefix itself, and the header becomes Bearer Bearer ….
6. setx does not affect the current window. It applies to windows opened afterwards. Set $env: as well to use it right away.
7. PowerShell curl. It is an alias for Invoke-WebRequest and does not accept these arguments. Use curl.exe.
8. An Anthropic account login appears. The variables are empty in this process. Close every terminal, including the ones inside your editor, and reopen from a shell where the variables are exported.

To use a model from a listing that isn't on this protocol, see When protocols don't match on the Install modelfold-chat-agents page.

Successful model discovery does not guarantee generation availability. For 429 Resources are tight or 400 Internal error, see the minimal check and troubleshooting guide. These responses do not count as a successful integration test.