Connector types
A named connection to an external system, and what every type of one carries.
Connector validation errors return 400 VALIDATION_ERROR with field paths.
Workflow execution failures that cannot be exposed return 500 ENGINE_ERROR,
while an open circuit returns 503 CIRCUIT_OPEN. Correct the connector definition,
test it through POST /connectors/{id}/test, and use Errors & Response
Envelopes for the complete contract.
A connector is a named connection to an external system — an API, a database, a cache, a Kafka cluster, or a search cluster. Workflows reference connectors by name. Credentials stay in the connector, never in workflow JSON.
There are exactly seven connector types:
| Type | Backs | Task functions |
|---|---|---|
http | REST APIs and webhooks | http_call |
kafka | Kafka topics (produce only) | publish_kafka |
db | PostgreSQL, MySQL, SQLite, MongoDB | data_query, data_write, db_read, db_write, mongo_read, mongo_write, mongo_aggregate |
cache | Redis or in-process memory | cache_read, cache_write |
es | Elasticsearch | data_query, data_write |
smtp | Transactional email over SMTP | send_email |
storage | S3-compatible object storage (presign + metadata only) | storage_presign, storage_head |
Type values match case-insensitively. Any other value is refused with the valid list. A stored connector whose config no longer parses fails to load and surfaces as a connector load issue.
The task functions and their inputs are specified in the Function Reference.
| Page | Holds |
|---|---|
| Shared connector blocks | identity, secret references, masking, authentication, request layering, gates and retries. |
| The seven types | the seven types, what each backs, and the config each takes. |
Related
- Admin API — Connectors: the endpoints, and the reachability probe.
- Task functions: the functions that call through a connector.
- Portable data dialect: the backend-neutral language
dbandesserve. - Environment variables:
${VAR}versusenv://, and what an unset variable does.
Last verified 14 September 2026