Creator data for Claude Code.
Connect creator discovery to Claude Code with a remote MCP server and an explicit request budget.
Connect your data key
Create a data key in the API console with prepaid credits or an active API subscription. Merge this server into your project .mcp.json. Set ROSTER_API_KEY in the environment that launches Claude Code. The placeholder expands locally; keep the actual key out of source control.
{
"mcpServers": {
"roster": {
"type": "http",
"url": "https://api.ugcroster.com/v1/mcp",
"headers": {
"Authorization": "Bearer ${ROSTER_API_KEY}"
}
}
}
}Check the connection
Restart Claude Code, open /mcp and check that roster connects. Ask for get_usage first to verify authentication without consuming data credits.
Make your first search
Find up to 20 US fitness creators with 10,000–100,000 followers. Use one search_creators call, page 1, limit 20. Budget: 10 credits, no profile reads, no retries or extra pages. Return a shortlist with available fields and state any missing data. Stop after the search, even if fewer than 20 creators match.
One search page costs 10 credits, whether it returns 1 or 20 matches. Each optional profile read costs 1 credit. Checking usage costs no credits. The prompt is an agent instruction, not a server-enforced spending cap; your application must enforce a job budget for unattended runs.
Work with the data you have
Results come from stored directory records. Location describes the creator, not their audience. An email on file is not a deliverability guarantee, and reading a profile does not trigger a live scrape. Reuse fields already returned by search before spending credits on extra profile reads.
Use a data key for these examples. It cannot send messages, create campaigns or record payouts. Do not enable brand actions for a discovery-only workflow.