KeyPool Integration Guide¶
KeyPool gives your team a shared API and tool surface for services with different SDKs, request formats and usage limits. Your token's granted access determines which services you can use.
What You Get¶
- One team token for the services granted to you.
- Shared service access through one stable endpoint per supported provider.
- Consistent failure handling when a provider is temporarily unavailable or rate-limited.
- Usage controls defined by your KeyPool workspace.
- Service-specific guidance for SDK setup, base URLs, and edge cases.
Choose how to connect¶
| Your task | Start here |
|---|---|
| Call a provider from an application | Follow the API/SDK steps below, then a service guide |
| Use tools in ChatGPT or Codex | MCP and plugin setup |
| Inspect endpoint parameters and responses | API reference |
API support and MCP tool availability have separate scopes. Check the MCP guide for the tools exposed to your client and the access granted to your connection.
Call an API or SDK¶
- Get your KeyPool Team Token from your workspace admin.
- Pick the service you want to integrate from the navigation.
- Set
KEYPOOL_BASE_URLto the base URL provided by your KeyPool administrator. - Send
Authorization: Bearer <TEAM_TOKEN>on every request to KeyPool. - Use the API Reference page to test the exact service endpoints in-browser.
Base URL Patterns¶
Most services use the same URL pattern:
https://your-keypool.example.com/v1/{service}
Two services use a service header for SDK compatibility:
- Firecrawl SDKs should target the KeyPool origin and use
x-keypool-service: firecrawl. - Tavily examples in this guide use the KeyPool origin plus
x-keypool-service: tavilyfor the same reason.
Availability And Errors¶
Inspect the HTTP status, JSON error and any Retry-After header. A 429 indicates
a usage/rate limit; a 502 or 503 means the request could not be completed.
A persistent 503 can require administrator action rather than repeated retries.
Retry only when the operation is safe to repeat and its service guide permits it.
Respect Retry-After, use bounded backoff, and avoid duplicating resource creation
or billable generation after an uncertain result. Contact your workspace admin
with the service, operation, status and request time when a failure persists.