Claude Code vs Windsurf

Claude Code and Windsurf solve the same problem — an agent that edits code from a task description instead of mouse clicks — but they sit at different levels of the stack. One is a CLI tool with no editor of its own; the other is a standalone IDE, a VS Code fork with a built-in agent called Cascade.

Where the agent lives

Claude Code runs in a terminal and works on top of any editor, or none at all: it opens the project, reads files, proposes edits as a diff, and runs commands on request. It doesn't care which editor you look at the result in afterward — VS Code, Vim, a JetBrains IDE — because it isn't an editor itself, just a process working against the filesystem and a terminal.

Windsurf is a separate application. The Cascade agent is built into the editor and sees open tabs, project structure, and edit history directly through the IDE interface, without needing the context re-explained each session. That's convenient when development happens in one window start to finish, but it means moving into a new application rather than adding a tool to an already-configured environment.

Connecting your own API

Claude Code is configured through environment variables — ANTHROPIC_BASE_URL and a key. One detail matters for anyone pointing it at a third-party gateway: the key variable is ANTHROPIC_AUTH_TOKEN, not ANTHROPIC_API_KEY — with the latter, Claude Code assumes you're using an official Anthropic account and prompts for login instead of using the key. Through Zukko it looks like this:

export ANTHROPIC_BASE_URL="https://api.zukko.pro"
export ANTHROPIC_AUTH_TOKEN="fetch_your_key"

Note the URL has no /v1 — as an Anthropic-compatible client, Claude Code appends the path itself and hits POST /v1/messages with the anthropic-version: 2023-06-01 header. A mixed-up suffix is the single most common cause of a 404 on first connection, and the symptom is distinctive: the key is accepted, but any model call returns 404.

Windsurf is structured differently here: connecting a third-party provider typically goes through the application's own settings rather than process environment variables, and how much customization is available depends on the app version and plan — part of the configuration stays inside the IDE interface rather than in a portable config you can copy between machines in one line. At the time of writing, Windsurf does not let you set your own base URL for all models — check the current BYOK settings in your version. If you want a VS Code-style editor with your own gateway, a working alternative is the Cline, Roo Code, or Kilo Code extensions in VS Code.

The model catalog and switching between them

At Zukko, one key opens several model lines at once: claude-opus-5-5 (Claude Opus 5.5), claude-sonnet-5, gpt-5.6-sol, gpt-5.6-luna, glm-5.3, and glm-5.3-flash (the current list is at GET https://api.zukko.pro/v1/models). For Claude Code running in Anthropic mode, these are model values like claude-sonnet-5 in the request body — switching between Opus and Sonnet needs no key or config change, just a different string in the call.

curl https://api.zukko.pro/v1/messages \
  -H "Authorization: Bearer fetch_your_key" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-opus-5-5","max_tokens":512,"messages":[{"role":"user","content":"check this module for a race condition"}]}'

Spend and control

Through Zukko, any connected tool hits the same billing: regular (non-streaming) responses carry X-Zukko-Cost-Micros (the request's cost in microdollars) and X-Fetch-Balance-Cents (the remaining balance after the charge, also in microdollars). But Claude Code works in streaming mode (stream: true), and streaming responses do not include these headers — you check the agent's spend in the dashboard (Usage), via GET /v1/usage, or GET /v1/balance. For an agent that can take several steps in a row without human input, this matters: a long task involving several files read and edited costs more than a single short question.

At the key level you can set a daily and monthly dollar spend limit and a list of allowed models, configured in the dashboard on the key's own page. That's particularly useful for agentic tools: if Claude Code is running in the background on CI to push through automated edits, a key-level limit protects against a runaway bill if something loops.

First-connection errors

Symptoms on both tools look similar, and diagnosis starts the same way: check the base URL and the key environment variable. A rejected key returns 401 with {"error":{"message":"Invalid API key. Use Authorization: Bearer fetch_*","type":"authentication_error"}}. An empty balance returns 402 insufficient_balance, checked before the request reaches the model — a failed request costs nothing. A misspelled model name gives 404 model_not_found; the current list without a key is at GET /v1/models.

Which to pick

If development is already built around a terminal and a familiar editor, and the goal is adding an agent without disrupting the rest of the workflow, that favors Claude Code: it layers on top rather than replacing anything. If it's more comfortable to work in one window where the agent sees editor context without extra explanation, that's an argument for Windsurf. Claude Code connects to a Zukko key directly; for Windsurf, first check whether your version lets you set your own base URL. The key is created in the dashboard at zukko.pro/app/api-keys↗ and shown once, at creation.