Developer documentation
How agents recover from auth, rate-limit, pagination, streaming, batch, webhook, and MCP errors.
For 401 or 403 responses, agents should stop retrying with the same key, surface the missing or invalid API key state, and direct the user to the API key and pricing docs. Discovery resources should remain available without browser-only authentication.
For 429 responses, agents should read Retry-After and RateLimit-Reset, wait, and retry only when the requested work is still useful. Agents should avoid parallel retries that consume the same budget.
For mutation requests, agents should include Idempotency-Key before the first attempt and reuse the same key only for the same operation. A 409 idempotency error means the key was reused for a different request.
BLERBZ feed and sync APIs use cursor pagination. Agents should persist the last successful cursor and resume from it after network failures instead of restarting long scans.
For SSE or streaming responses, agents should keep the last complete event, reconnect only when useful, and prefer a shorter query or smaller result set if repeated disconnects occur.
If MCP initialize fails, verify /.well-known/mcp.json, /.well-known/mcp/server-card.json, the /mcp serverUrl, protocolVersion, initialize, tools/list, resources/list, and resources/read. Public discovery should not require browser-only auth.