Build Custom Connectors with AI
Any REST API, SOAP service, GraphQL endpoint, or JDBC database can become a Supaflow connector. An open-source skill framework guides AI coding agents through an 8-phase build process with gate checks, anti-pattern enforcement, and automated verification.
Works with Claude Code, OpenAI Codex, and any agent that supports skill or prompt files. The agent writes the code. The skill ensures it meets production standards.
What You Can Build
The SDK and skill framework support three connector modes. Each follows a tailored phase track with mode-specific verification checks.
Source connectors
Extract data from REST APIs, SOAP services, GraphQL endpoints, or JDBC databases. Supports full refresh, incremental sync, and re-import modes.
Warehouse destinations
Load data into warehouses like Snowflake or BigQuery via staging and bulk load operations. Handles DDL generation, MERGE for upserts, and schema evolution.
Activation targets
Push data from your warehouse back to SaaS platforms like Salesforce or HubSpot via reverse ETL. Maps warehouse columns to destination object fields.
How It Works
The AI agent follows a phased workflow. Each phase has prerequisite reading, a defined scope, and a gate check that must pass before proceeding to the next phase.
Project setup and connector identity
The agent scaffolds the Maven module, pom.xml with shade plugin, connector shell class, and registers connector properties and capabilities (source, destination, or both).
Connection and authentication
Implements the connection layer with support for OAuth 2.0 (auth code or client credentials), API keys, basic auth, or JDBC. Includes token refresh, connection pooling, and credential validation.
Schema discovery and read operations
Builds ObjectMetadata and FieldMetadata with proper type mapping and originalDataType tracking. Implements read operations with pagination, cursor-based incremental sync via CutoffTimeSyncUtils, and cancellation checks in long-running loops.
Integration testing and verification
Writes and runs integration tests against a live API. A bundled verification script runs 24 automated checks covering processor lifecycle, field metadata, capability declarations, incremental sync, naming conventions, and build artifacts.
SDK Features
Production-grade connectors, not prototypes. The Java SDK and Python SDK provide the building blocks, and the skill framework enforces quality.
Incremental sync
Track cursor positions per object with CutoffTimeSyncUtils so subsequent runs fetch only changed records.
Schema discovery
Auto-discover objects, fields, and types from the source API. Set primary keys, cursor fields, and sync modes per object.
Error handling
Built-in retry logic, connection validation, and cancellation checks in every long-running loop.
Type mapping
Map source types to Supaflow canonical types (BOOLEAN, INT, LONG, DOUBLE, STRING, JSON, INSTANT, etc.) with safe widening.
Anti-pattern enforcement
20+ known anti-patterns are caught by gate checks before code reaches production -- from processor lifecycle violations to missing cursor fields.
Automated verification
24 checks validate the connector end-to-end: build config, capabilities, auth, schema discovery, read operations, and integration tests.
Get the SDK
Java SDK (Maven)
Add the dependency to your connector module. The SDK provides the SupaflowConnector interface, base classes for source and destination connectors, and the full pipeline lifecycle API.
<groupId>io.supa-flow</groupId>
<artifactId>supaflow-connector-sdk</artifactId>View on Maven CentralPython SDK (pip)
Build source connectors in Python with built-in support for schema discovery, incremental sync, and type mapping.
pip install supaflow-python-sdkOpen Source
The Connector Dev Skills are open source and available on GitHub. Clone the repo, symlink the skill into your agent, and start building. The symlink points to the live repo so updates are picked up with a simple git pull.
Back to Connectors