Use your Claude subscription in Tarsk
Sign in with your Claude account and run Tarsk chats on Opus, Sonnet and Haiku, billed to your plan instead of per token.
If you pay for a Claude plan, you can now use it in Tarsk. The new Claude provider signs in with your Claude account, and your chats count against your subscription instead of a pay-per-token API key.
The Claude provider sits next to the Anthropic provider in the list. Anthropic still takes an API key and bills per token. Claude takes a sign-in and bills your plan.
Connect your account
- Open Settings, then Providers, and pick Claude.
- Click Connect Claude. Tarsk opens the Claude sign-in page in your browser.
- Approve the sign-in. Claude shows you an authorization code.
- Paste it into Paste the authorization code from the browser and click Finish sign-in.
The page then reads Signed in on this machine as Claude, with Reconnect and Disconnect buttons.
Tarsk does not store your Claude credential. The sign-in lives in Claude’s own sign-in store on that machine, and Tarsk asks it whether you are signed in. A remote machine that runs your chats needs its own sign-in, done from that machine’s provider page.
Tarsk ships the Claude runtime it needs, so you do not have to install anything. If you already use Claude Code from the terminal, Tarsk also finds it in its default install location.
Pick a model
When you connect, Tarsk asks Claude which models your account offers and turns on the current ones, such as Claude Opus, Claude Sonnet and Claude Haiku. Pick one in the model selector like any other model. Claude Sonnet is the default.
The reasoning setting you choose for a model applies to Claude runs too. Set a higher effort for a hard refactor and a lower one for quick questions.
Tarsk stays in charge of the tools
A Claude chat behaves like any other Tarsk chat. Claude runs the turn, and Tarsk supplies the tools, the project rules, the enabled skills and the system prompt. Your MCP servers, run_js and tool permissions work the same way. Claude’s own built-in tools stay off, so each file edit and shell command goes through the permission rules you set in Tarsk.
Stop works as usual and ends the run and its tool calls. When you come back to a chat, Tarsk continues the same Claude session, so Claude keeps the context of earlier turns.
You can switch a chat between Claude and another provider. Tarsk passes the messages that the other model wrote to Claude when you switch back.
Watch your plan, and what it saves
A subscription limits your usage per time window. The context usage panel of a Claude chat has a Plan limits section with each window and its reset time, such as the 5-hour limit and the Weekly limit. If you reach a limit, the chat pauses with a message and you can retry later.
The spend widget adds an API equivalent line for the current conversation. It shows what the same tokens would cost at pay-per-token API prices, so you can see what the plan saves you.
Try it
Connect Claude in Settings → Providers, pick Claude Sonnet in a project you know well, and give it a task you would usually give an API model. Then open the spend widget and compare the API equivalent with your plan price.
- Sign in once per machine with your Claude account.
- Claude runs the turn, and Tarsk keeps its tools, skills and rules.
- Plan limits and the API equivalent show you where you stand.