1. Get your key
Sign in with GitHub. Your personal boost page and its key are created on first sign-in, and the key is shown once on your dashboard. It is a boostmyagents virtual key, not an API key issued by Anthropic, and it only works against the boostmyagents gateway.
2. Set two environment variables
Anthropic-compatible clients read the gateway address from ANTHROPIC_BASE_URL and the key from ANTHROPIC_API_KEY.
export ANTHROPIC_BASE_URL=https://api.boostmyagents.com
export ANTHROPIC_API_KEY=your-virtual-key
Your dashboard shows these two lines with your real key filled in when the key is created. To keep them across terminal sessions, add them to your shell profile, such as ~/.zshrc or ~/.bashrc.
3. Run Claude Code
Start Claude Code from the same terminal. It picks up both variables automatically.
claude
To choose a specific model, pass one of the supported model IDs:
claude --model claude-sonnet-5
4. Test the connection
This sends one small request through the gateway. A JSON reply with a content field means the key and address are working.
curl "$ANTHROPIC_BASE_URL/v1/messages" \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-haiku-4-5-20251001",
"max_tokens": 32,
"messages": [{"role": "user", "content": "Say hello"}]
}'
Using the Anthropic SDK
The official SDKs read the same two variables, so no code change is needed beyond choosing a supported model.
from anthropic import Anthropic
client = Anthropic() # reads ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY
message = client.messages.create(
model="claude-sonnet-5",
max_tokens=1024,
messages=[{"role": "user", "content": "Summarize this changelog."}],
)
print(message.content[0].text)
Supported models
The gateway accepts only these model IDs. Requests for any other model are rejected and not charged.
claude-fable-5-1claude-fable-5claude-opus-5-5claude-opus-4-8claude-opus-4-7claude-opus-4-6claude-sonnet-5claude-sonnet-4-6claude-haiku-4-5-20251001
A key can use the models that were on this list when the key was created. If a model is added later, regenerate your key to use it.
Gateway limits
The gateway only accepts requests whose cost it can calculate before they are sent, so a boost can never be overspent.
- One request at a time per key.
- Only the
/v1/messagesendpoint is available. - Tools that run on your own machine work normally, including Claude Code’s file, search, and shell tools.
- Tools that run on Anthropic’s servers are rejected: hosted web search (which Claude Code’s built-in web search relies on), web fetch, code execution, and hosted MCP servers.
- Fast mode, priority tiers, server-side model fallbacks, and server-side conversation compaction are rejected.
- The gateway may cap a request’s maximum output length so it fits the remaining boost.
- Requests that would cost more than what remains are rejected before they reach Anthropic.
Rejected requests are not charged.
Keeping your key safe
Treat the key like a password: keep it out of repositories, CI logs, and screenshots. If it leaks or you lose it, regenerate it from your dashboard and update ANTHROPIC_API_KEY.