Claude Code
Claude Code talks to MartLLM through its Anthropic-compatible address,
https://api.martllm.com. With the address and your key set, Claude Code's
commands and tools work as before, and requests are billed to your MartLLM
balance.
Set it up by hand
Nothing to run but Claude Code itself. Make a key on your keys page, then:
export ANTHROPIC_BASE_URL=https://api.martllm.com
export ANTHROPIC_AUTH_TOKEN=ull_your_key
claude
Windows PowerShell:
$env:ANTHROPIC_BASE_URL = "https://api.martllm.com"
$env:ANTHROPIC_AUTH_TOKEN = "ull_your_key"
claude
That lasts for the current terminal. To keep it without a key in your shell profile, save the key to a file only you can read:
mkdir -p ~/.uberllm && chmod 700 ~/.uberllm
printf '%s\n' 'ull_your_key' > ~/.uberllm/key && chmod 600 ~/.uberllm/key
and put this in ~/.claude/settings.json (every project) or in
<project>/.claude/settings.local.json (one project):
{
"apiKeyHelper": "cat ~/.uberllm/key",
"env": {
"ANTHROPIC_BASE_URL": "https://api.martllm.com"
}
}
Or use the installer
If you have Node.js, one command sets up Claude Code and the other coding tools on your machine (Codex CLI, OpenCode and more) with the same settings as the installer below:
npx uberllm init
See uberllm init for what it changes and how to undo it.
The installer does the same thing for you and gets the key through your browser. Read it first if you like: it is plain text at martllm.com/install/claude-code.sh.
- SHA-256 of
claude-code.sh:34646d5a83e25a1c19057a6bca6aeaaf0118e0e43704d5dbb01e49ab15eb625f - SHA-256 of
claude-code.ps1:5caaa5f15f3e846c535e6668687fe9e0bdcf1e4f9a8a370af09a5137005f44b4
To check it before running it:
curl -fsSL https://martllm.com/install/claude-code.sh -o claude-code.sh
shasum -a 256 claude-code.sh # compare with the hash above
sh claude-code.sh
Every project on this computer:
curl -fsSL https://martllm.com/install/claude-code.sh | sh
One project only (run it inside the project; other projects stay on your Anthropic plan):
curl -fsSL https://martllm.com/install/claude-code.sh | sh -s -- --project
Add a folder after --project to set up a project other than the current one.
Windows PowerShell (every project):
irm https://martllm.com/install/claude-code.ps1 | iex
You do not need an API key first, and you do not need to be signed in when you
paste the command. The installer opens a link in your browser
(https://martllm.com/cli/link?code=XXXX-XXXX) and prints the same confirmation
code in the terminal. Sign in, check the code matches, and press Approve.
Approving creates an API key on your account and hands it to the waiting installer. The code in your terminal is not the secret: only the installer that started the request can collect the key, the key is handed out exactly once, and the request expires after 15 minutes.
If Claude Code has never been set up on this computer, use the every-project install first. Claude Code reads a project's own settings only after you trust that folder, so a brand-new install could still ask you to log in.
To print the settings block for your computer without writing anything:
curl -fsSL https://martllm.com/install/claude-code.sh | sh -s -- --print
What the installer writes
The key goes into a file only you can read (folder mode 700, file mode 600):
| Install | Key file | Settings file |
|---|---|---|
| Every project | ~/.uberllm/key |
~/.claude/settings.json (or $CLAUDE_CONFIG_DIR/settings.json) |
| One project | ~/.uberllm/key-<folder>-<number> |
<project>/.claude/settings.local.json |
The settings file gets the address and an apiKeyHelper that reads the key file.
The key itself is never written into a settings file, so it does not end up in a
backup, a dotfiles repository or a screen share.
{
"apiKeyHelper": "cat \"/Users/you/.uberllm/key\"",
"env": {
"ANTHROPIC_BASE_URL": "https://api.martllm.com",
"CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY": "1"
}
}
Claude Code runs apiKeyHelper to get the key and sends it as both
Authorization: Bearer and x-api-key, so it works with a gateway such as
MartLLM. It keeps the answer for five minutes (CLAUDE_CODE_API_KEY_HELPER_TTL_MS
changes that).
The installer merges into your existing file rather than replacing it, so hooks,
permissions and MCP servers are kept, and it copies the old file to
<file>.bak-<timestamp> first. It removes a plain-text ANTHROPIC_AUTH_TOKEN
left by an older install. It then makes one real request through the gateway, so
a success message means a request went through.
The Windows installer still writes the key into settings.json as
ANTHROPIC_AUTH_TOKEN.
Which setting wins
Claude Code reads settings in this order, first one wins:
- Managed settings (set by your company, if any)
claude --settings <file><project>/.claude/settings.local.json<project>/.claude/settings.json~/.claude/settings.json
A variable in a settings file's env block wins over the same variable in your
shell in most sessions. Two exceptions matter here:
- The key. Claude Code uses
ANTHROPIC_AUTH_TOKENfirst, thenANTHROPIC_API_KEY, thenapiKeyHelper. A token exported in your shell replaces the saved key. - The model.
/modelin the session wins, thenclaude --model, thenANTHROPIC_MODEL, then themodelsetting. A shellANTHROPIC_MODELbeats themodelsetting.
So remove every ANTHROPIC_* variable from your shell and your shell profile
(~/.zshrc, ~/.bashrc) and keep them in the settings file. The installer warns
you when it finds one.
Use other models
By default nothing is pinned: Claude Code asks for its own model names (opus,
sonnet, haiku, claude-opus-5-5…) and MartLLM serves the same Claude models.
To drive Claude Code with DeepSeek, GLM, Qwen, GPT or any other model, map each
of Claude Code's model slots to a MartLLM model id in the env block:
| Variable | What it sets |
|---|---|
ANTHROPIC_MODEL |
The model the session starts on. |
ANTHROPIC_DEFAULT_OPUS_MODEL |
What opus means. |
ANTHROPIC_DEFAULT_SONNET_MODEL |
What sonnet means. |
ANTHROPIC_DEFAULT_HAIKU_MODEL |
What haiku means. Claude Code also uses it for background work such as titles and summaries. |
CLAUDE_CODE_SUBAGENT_MODEL |
The model subagents run on. |
{
"apiKeyHelper": "cat \"/Users/you/.uberllm/key\"",
"env": {
"ANTHROPIC_BASE_URL": "https://api.martllm.com",
"ANTHROPIC_MODEL": "deepseek/deepseek-v4-pro",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek/deepseek-v4-pro",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek/deepseek-v4-pro",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek/deepseek-v4-flash",
"CLAUDE_CODE_SUBAGENT_MODEL": "deepseek/deepseek-v4-pro"
}
}
Map every slot. A slot left unset keeps asking for a Claude model, and you pay Claude prices for it.
The installer does this for you with a profile:
curl -fsSL 'https://martllm.com/install/claude-code.sh?profile=glm' | sh
Profiles are claude-native (the default, pins nothing), glm, deepseek and
gpt. Any model id from the models table works if
you set the variables by hand; GET https://api.martllm.com/v1/models lists them.
Context and compaction
Claude Code knows the context size of Claude models only. For any other model it assumes 200,000 tokens and compacts the conversation before that.
- If the model has a 1-million-token context (its page on martllm.com
shows the context), add
[1m]to the end of its id, for example"ANTHROPIC_MODEL": "deepseek/deepseek-v4-pro[1m]"orclaude --model 'deepseek/deepseek-v4-pro[1m]'. Claude Code then plans for 1M. - For any other size, set
CLAUDE_CODE_MAX_CONTEXT_TOKENSto the model's context, for example"CLAUDE_CODE_MAX_CONTEXT_TOKENS": "400000". - To choose when it compacts: the
autoCompactWindowsetting (or theCLAUDE_CODE_AUTO_COMPACT_WINDOWvariable) takes a number of tokens from 100,000 to 1,000,000.CLAUDE_AUTOCOMPACT_PCT_OVERRIDE(1–100) compacts earlier, at that share of the window.autoCompactEnabled: falseturns it off.
Setting a context larger than the model has makes requests fail once the conversation outgrows the model.
What changes
- Your claude.ai login is not used. Features that need it, such as claude.ai
connectors,
/scheduleand Remote Control, are not available in a session that runs on MartLLM. - The cost Claude Code shows is not your bill. Claude Code works it out on your computer from Anthropic's list prices, and shows nothing useful for other models. What you pay is in Dashboard → API Keys → API Usage. Add your balance to the status line (below) to see it while you work.
- Your code goes to third-party model providers. Everything the session reads is sent to the model, which is run by third-party model providers on the MartLLM market. Do not use it in a repository that holds secrets, credentials or customer data.
Balance in the status line
curl -fsSL https://martllm.com/install/statusline.sh | sh -s -- --install
This saves a small script to ~/.uberllm/statusline.sh and adds it as Claude
Code's statusLine in ~/.claude/settings.json (add --project after --install
to put it in this project's .claude/settings.local.json instead). Under the
prompt you then see:
MartLLM $12.40 · $0.35 held · Opus
That is your available balance, the amount held for requests still running, and
the model. It reads GET /v1/balance with the same key Claude Code uses, at most
once a minute. It needs only sh, curl and sed. Read it first at
martllm.com/install/statusline.sh.
Verify
Start Claude Code and run /status: the API address is
https://api.martllm.com. Then ask it anything; the request shows in
Dashboard → API Keys → API Usage within a minute.
The same key answers over the Messages API:
curl https://api.martllm.com/v1/messages \
-H "Authorization: Bearer $(cat ~/.uberllm/key)" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{"model":"claude-haiku-4-5","max_tokens":16,"messages":[{"role":"user","content":"ping"}]}'
If a request fails with Your apiKeyHelper script is failing, the key file is
missing or unreadable: run the installer again.
Uninstall and revoke
curl -fsSL https://martllm.com/install/claude-code.sh | sh -s -- --uninstall
For a one-project install, run it inside the project with
sh -s -- --project --uninstall. It backs up the settings file first, removes
only what it added, and deletes the saved key file. Claude Code goes back to your
Anthropic login on the next start.
Uninstalling does not revoke the key; it only stops using it. Delete the key on your keys page to make sure it can never spend again. Do that for any computer you no longer trust, or if you approved a request you did not start.