TokelyTHE SHORT ANSWER
Claude Code can use Tokely through its Anthropic-compatible Messages endpoint. Configure the gateway base URL, a Tokely credential and supported Claude model IDs. Compare the full session cost, including cache writes, cache reads and output, before assuming that a lower headline input price makes coding cheaper.
Configuration and documentation checked October 15, 2026.
Which endpoint does Claude Code need?
Claude Code does not become compatible with a provider merely because that provider supports OpenAI Chat Completions. The gateway must implement a supported Claude Code API format. Tokely documents a native /v1/messages endpoint with streaming, tool messages and prompt caching. Tokely Claude Code guide.
Anthropic’s gateway documentation explains the integration model and its support boundary. Routing to third-party gateways does not imply that Anthropic endorses them.
How do you configure Tokely?
With Claude Code already installed and a funded Tokely account, start it with this single-line shell command. Replace the key placeholder with your Tokely key:
ANTHROPIC_BASE_URL=https://api.tokely.me ANTHROPIC_AUTH_TOKEN='<YOUR_TOKELY_API_KEY>' ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-5 ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-5-5 ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-5-5 claudeUse the base URL exactly as shown, without appending /v1/messages. These model mappings follow Tokely’s published guide at the check date. Confirm that the models remain enabled before using the configuration later. Inspect existing Claude Code settings for conflicting endpoint or authentication configuration.
Does a Claude subscription pay for these requests?
Requests authenticated with the gateway credential are billed through that gateway arrangement. A Claude subscription is not a general pool of API credits you can spend at Tokely. The official gateway billing explanation distinguishes gateway credentials from subscription authentication.
Why is caching especially relevant to coding?
Coding sessions repeatedly send project instructions, tool definitions and conversation context. Record how much input qualifies as cache reads and how much requires new writes. A session with frequent prefix changes can cost more than one with stable context, even if both edit the same number of files.
Compare total cost per accepted code change. Include retries, failed attempts and time spent repairing generated code. A cheaper model that needs several extra attempts may not be the cheaper workflow.
What limitations should you know before switching?
Tokely’s documented Messages surface excludes /v1/messages/count_tokens, the Batch API, Files API and Anthropic server tools such as web search. This means the endpoint should not be described as complete parity with every Anthropic API feature. Supported features and limits.
Use ordinary repository tasks to evaluate compatibility when you choose to run a trial. Keep the model mapping and session usage together so a later price comparison is reproducible.
Review Tokely’s Claude prices, create a key through the quickstart, and use the documented endpoint for the integration.