External connectors

Integrate external data sources (MySQL, PostgreSQL, SQL Server, REST APIs) alongside Cosmos DB. Connector models use the same Zod-driven scaffold workflow and produce identical BFF routes — the frontend does not need to know where data comes from.

A connector model is a standard Zod model with an additional connectorConfig export that tells SwallowKit how to reach the external data source.

Architecture

Connector models plug into the same architecture as standard Cosmos DB models. The frontend and BFF layer are completely unaware of the underlying data source — the difference is only in the Functions layer.

Standard Model (Cosmos DB)

Frontend → BFF (Next.js API Routes) → Azure Functions → Cosmos DB

Connector Model

Frontend → BFF (Next.js API Routes) → Azure Functions → External Source (RDB / API)

The BFF routes generated for connector models are identical to standard models. This means:

The frontend consumes all models through the same interface
You can migrate a model from Cosmos DB to an external source (or vice versa) without changing the frontend
UI components are generated the same way regardless of the data source

Local Development with Mock Connectors

Frontend → BFF → Mock Proxy (:7072) ──┬─ Connector routes → In-memory CRUD (Zod-generated data)
                                      └─ Other routes → Azure Functions (:7071)

The mock proxy intercepts requests to connector models and serves realistic fake data, while proxying standard Cosmos DB model requests to the real Azure Functions runtime.

Getting Started

1. Add a Connector to Config

Use the add-connector command to register a new external data source:

npmpnpm

bash

# Add an RDB connector (MySQL)
npx swallowkit add-connector mysql --type rdb --provider mysql

# Add an API connector (Backlog)
npx swallowkit add-connector backlog --type api

bash

# Add an RDB connector (MySQL)
pnpm exec swallowkit add-connector mysql --type rdb --provider mysql

# Add an API connector (Backlog)
pnpm exec swallowkit add-connector backlog --type api

This adds entries to the connectors section of your swallowkit.config.js. You can also edit the config file manually — see the Configuration Reference below.

2. Create a Connector Model

Use create-model with the --connector flag to generate a model template that includes connectorConfig:

npmpnpm

bash

# Create a model bound to the mysql connector
npx swallowkit create-model user --connector mysql

# Create a model bound to the backlog connector
npx swallowkit create-model backlog-issue --connector backlog

bash

# Create a model bound to the mysql connector
pnpm exec swallowkit create-model user --connector mysql

# Create a model bound to the backlog connector
pnpm exec swallowkit create-model backlog-issue --connector backlog

This generates shared/models/user.ts (or backlog-issue.ts) with a Zod schema and a connectorConfig export pre-filled for the specified connector.

3. Edit the Model

Customize the generated Zod schema and connectorConfig to match your actual data source:

RDB example (shared/models/user.ts):

typescript

import { z } from 'zod/v4';

export const User = z.object({
  id: z.string(),
  name: z.string(),
  email: z.string().email(),
  department: z.string().optional(),
  createdAt: z.string().optional(),
  updatedAt: z.string().optional(),
});

export type User = z.infer<typeof User>;
export const displayName = 'User';

export const connectorConfig = {
  connector: 'mysql',
  operations: ['getAll', 'getById'] as const,
  table: 'users',
  idColumn: 'id',
};

API example (shared/models/backlog-issue.ts):

typescript

import { z } from 'zod/v4';

export const BacklogIssue = z.object({
  id: z.string(),
  projectId: z.string(),
  issueKey: z.string(),
  summary: z.string(),
  description: z.string().optional(),
  status: z.object({ id: z.number(), name: z.string() }),
  assignee: z.object({ id: z.number(), name: z.string() }).optional(),
  priority: z.object({ id: z.number(), name: z.string() }).optional(),
  createdAt: z.string().optional(),
  updatedAt: z.string().optional(),
});

export type BacklogIssue = z.infer<typeof BacklogIssue>;
export const displayName = 'BacklogIssue';

export const connectorConfig = {
  connector: 'backlog',
  operations: ['getAll', 'getById', 'create', 'update'] as const,
  endpoints: {
    getAll: 'GET /issues',
    getById: 'GET /issues/{id}',
    create: 'POST /issues',
    update: 'PATCH /issues/{id}',
  },
};

4. Scaffold

Run scaffold as usual — SwallowKit detects the connectorConfig export and generates connector-specific Functions code instead of Cosmos DB code:

npmpnpm

bash

npx swallowkit scaffold shared/models/user.ts
npx swallowkit scaffold shared/models/backlog-issue.ts

bash

pnpm exec swallowkit scaffold shared/models/user.ts
pnpm exec swallowkit scaffold shared/models/backlog-issue.ts

Generated files depend on the backend language:

Component	Standard Model	Connector Model
Functions (C#)	`functions/`	`functions/Connectors/`
Functions (TypeScript)	`functions/src/functions/`	`functions/src/functions/`
Functions (Python)	`functions/`	`functions/`
BFF routes	Same	Same (transparent to frontend)
UI components	Same	Same
Cosmos DB Bicep	Generated	Skipped

⚠️ Read-only models: If operations only includes getAll and/or getById, scaffold will not generate POST, PUT, or DELETE handlers. The model is treated as read-only.

5. Local Development

Start the dev server with the --mock-connectors flag to work without real external connections:

npmpnpm

bash

npx swallowkit dev --mock-connectors

bash

pnpm exec swallowkit dev --mock-connectors

This starts the mock proxy server on port 7072 alongside the normal dev environment (Cosmos Emulator + Azure Functions on 7071 + Next.js). See Mock Server for details.

Configuration Reference

External data sources are defined in the connectors section of swallowkit.config.js. Each key is a unique connector name used in model connectorConfig.

Full Config Example

javascript

// swallowkit.config.js
module.exports = {
  backend: { language: 'csharp' },
  functions: {
    baseUrl: process.env.BACKEND_FUNCTIONS_BASE_URL || 'http://localhost:7071',
  },
  connectors: {
    mysql: {
      type: 'rdb',
      provider: 'mysql',
      connectionEnvVar: 'MYSQL_CONNECTION_STRING',
    },
    postgres: {
      type: 'rdb',
      provider: 'postgres',
      connectionEnvVar: 'POSTGRES_CONNECTION_STRING',
    },
    backlog: {
      type: 'api',
      baseUrlEnvVar: 'BACKLOG_API_BASE_URL',
      auth: {
        type: 'apiKey',
        envVar: 'BACKLOG_API_KEY',
        placement: 'query',
        paramName: 'apiKey',
      },
    },
    internal: {
      type: 'api',
      baseUrlEnvVar: 'INTERNAL_API_BASE_URL',
      auth: {
        type: 'bearer',
        envVar: 'INTERNAL_API_TOKEN',
      },
    },
  },
};

RDB Connector Options

Property	Type	Required	Description
`type`	`'rdb'`	✅	Connector type
`provider`	`'mysql'` \| `'postgres'` \| `'sqlserver'`	✅	Database provider
`connectionEnvVar`	`string`	✅	Environment variable name holding the connection string

API Connector Options

Property	Type	Required	Description
`type`	`'api'`	✅	Connector type
`baseUrlEnvVar`	`string`	✅	Environment variable name holding the API base URL
`auth`	`object`	✅	Authentication configuration (see below)

API Auth Types

API Key (type: 'apiKey'):

Property	Type	Required	Description
`type`	`'apiKey'`	✅	Auth type
`envVar`	`string`	✅	Environment variable holding the API key
`placement`	`'query'` \| `'header'`	✅	Where to send the API key
`paramName`	`string`	✅	Query parameter or header name

Bearer Token (type: 'bearer'):

Property	Type	Required	Description
`type`	`'bearer'`	✅	Auth type
`envVar`	`string`	✅	Environment variable holding the bearer token

OAuth 2.0 (type: 'oauth2'):

Property	Type	Required	Description
`type`	`'oauth2'`	✅	Auth type
`tokenUrlEnvVar`	`string`	✅	Environment variable holding the token endpoint URL
`clientIdEnvVar`	`string`	✅	Environment variable holding the client ID
`clientSecretEnvVar`	`string`	✅	Environment variable holding the client secret

Model Metadata Reference

The connectorConfig export in a model file tells SwallowKit which connector to use and how to interact with the external data source.

Common Fields

Property	Type	Required	Description
`connector`	`string`	✅	Name of the connector (must match a key in `connectors` config)
`operations`	`readonly string[]`	✅	Array of enabled operations

RDB-Specific Fields

Property	Type	Required	Description
`table`	`string`	✅	Database table name
`idColumn`	`string`	✅	Primary key column name

typescript

export const connectorConfig = {
  connector: 'mysql',
  operations: ['getAll', 'getById', 'create', 'update', 'delete'] as const,
  table: 'products',
  idColumn: 'id',
};

API-Specific Fields

Property	Type	Required	Description
`endpoints`	`object`	✅	Mapping of operation names to HTTP method + path

typescript

export const connectorConfig = {
  connector: 'backlog',
  operations: ['getAll', 'getById', 'create'] as const,
  endpoints: {
    getAll: 'GET /issues',
    getById: 'GET /issues/{id}',
    create: 'POST /issues',
  },
};

Endpoint paths use {id} as a placeholder for the resource identifier.

Supported Operations

Operation	HTTP Method	RDB (Generated SQL)	API (Generated HTTP)	Description
`getAll`	GET	`SELECT * FROM {table}`	`GET {endpoint}`	Retrieve all records
`getById`	GET	`SELECT * FROM {table} WHERE {idColumn} = ?`	`GET {endpoint}/{id}`	Retrieve a single record
`create`	POST	`INSERT INTO {table} ...`	`POST {endpoint}`	Create a new record
`update`	PUT	`UPDATE {table} SET ... WHERE {idColumn} = ?`	`PATCH {endpoint}/{id}`	Update an existing record
`delete`	DELETE	`DELETE FROM {table} WHERE {idColumn} = ?`	`DELETE {endpoint}/{id}`	Delete a record

💡 Read-only pattern: To create a read-only integration, only include getAll and/or getById in the operations array. Scaffold will skip write handlers, and the mock server will return 405 for write requests.

Mock Server

The --mock-connectors flag on the dev command starts a mock proxy server that eliminates the need for real external connections during development.

How It Works

Startup: The mock proxy starts on port 7072 and the BFF is configured to route through it
Routing: Requests for connector models are intercepted and handled in-memory; all other requests are proxied to Azure Functions on port 7071
Data generation: On startup, the mock server reads each connector model's Zod schema and generates realistic fake data using field-name heuristics (e.g., email fields get email-like values, name fields get name-like values)
In-memory CRUD: Generated data is stored in memory and supports full CRUD operations matching each model's operations array

Mock Data Generation

The mock server auto-generates realistic data based on Zod schema field names and types:

Field Pattern	Generated Value
`email`	Email address (e.g., `user@example.com`)
`name`	Person or entity name
`url`, `website`	URL string
`phone`	Phone number
`id`	UUID string
`createdAt`, `updatedAt`	ISO 8601 timestamp
`z.number()`	Random number within constraints
`z.boolean()`	Random boolean
`z.enum([...])`	Random value from the enum

Dev Seeds Integration

You can provide initial data for connector models using dev-seeds JSON files:

npmpnpm

bash

npx swallowkit create-dev-seeds shared/models/user.ts

bash

pnpm exec swallowkit create-dev-seeds shared/models/user.ts

This creates a JSON seed file that the mock server loads as initial data instead of auto-generated data. The seed file format is the same as for standard Cosmos DB models.

Unsupported Operations

If a request targets an operation not listed in the model's operations array, the mock server returns 405 Method Not Allowed.

# If user model only has ['getAll', 'getById']:
POST /api/users → 405 Method Not Allowed
DELETE /api/users/123 → 405 Method Not Allowed

Best Practices

When to Use Connectors

✅ Your project needs data from an existing relational database
✅ You want to integrate a third-party SaaS API into your SwallowKit app
✅ You need a read-only view of external data alongside Cosmos DB data
✅ You want consistent type-safe UI and BFF layer across all data sources

Read-Only Pattern

For external data sources you should not modify (e.g., a shared corporate database), use a read-only connector:

typescript

export const connectorConfig = {
  connector: 'mysql',
  operations: ['getAll', 'getById'] as const,
  table: 'employees',
  idColumn: 'employee_id',
};

This ensures:

Only GET endpoints are generated
No write-related UI components (create/edit forms) are generated
The mock server rejects write attempts with 405

Naming Conventions

Connector names in config: Use lowercase, descriptive names (e.g., mysql, backlog, salesforce)
Model files: Use kebab-case matching the model name (e.g., backlog-issue.ts)
Environment variables: Use UPPER_SNAKE_CASE with a clear prefix (e.g., BACKLOG_API_BASE_URL, MYSQL_CONNECTION_STRING)

Environment Variable Management

Store all connector credentials in environment variables — never hardcode connection strings or API keys:

bash

# .env.local (for local development)
MYSQL_CONNECTION_STRING=mysql://user:pass@localhost:3306/mydb
BACKLOG_API_BASE_URL=https://your-space.backlog.com/api/v2
BACKLOG_API_KEY=your-api-key-here

Mixing Cosmos DB and Connector Models

You can freely mix standard Cosmos DB models and connector models in the same project. The frontend and BFF layer treat them identically — only the Functions layer differs.

Limitations

The following are current limitations of the connector feature:

No caching layer: Connector-generated Functions do not include caching. Add caching manually if needed for high-traffic external API calls
No custom auth flows: OAuth 2.0 supports only the client credentials flow. Interactive/authorization code flows are not supported
No pagination pass-through: The getAll operation does not pass pagination parameters to the external source. Implement pagination manually for large datasets
No JOIN / relations: RDB connectors operate on a single table per model. Cross-table queries require custom implementation
No schema migration: SwallowKit does not create or migrate database tables for RDB connectors — tables must already exist
No webhook / event support: Connectors are request-response only; push-based integrations are not supported
No connection pooling config: RDB connection pooling relies on the driver defaults; advanced pool configuration is not exposed

💡 Reference: For CLI command details, see the CLI Reference. For standard Cosmos DB model scaffolding, see the Scaffold Guide.

External connectors ​

Architecture ​

Standard Model (Cosmos DB) ​

Connector Model ​

Local Development with Mock Connectors ​

Getting Started ​

1. Add a Connector to Config ​

2. Create a Connector Model ​

3. Edit the Model ​

4. Scaffold ​

5. Local Development ​

Configuration Reference ​

Full Config Example ​

RDB Connector Options ​

API Connector Options ​

API Auth Types ​

Model Metadata Reference ​

Common Fields ​

RDB-Specific Fields ​

API-Specific Fields ​

Supported Operations ​

Mock Server ​

How It Works ​

Mock Data Generation ​

Dev Seeds Integration ​

Unsupported Operations ​

Best Practices ​

When to Use Connectors ​

Read-Only Pattern ​

Naming Conventions ​

Environment Variable Management ​

Mixing Cosmos DB and Connector Models ​

Limitations ​

External connectors

Architecture

Standard Model (Cosmos DB)

Connector Model

Local Development with Mock Connectors

Getting Started

1. Add a Connector to Config

2. Create a Connector Model

3. Edit the Model

4. Scaffold

5. Local Development

Configuration Reference

Full Config Example

RDB Connector Options

API Connector Options

API Auth Types

Model Metadata Reference

Common Fields

RDB-Specific Fields

API-Specific Fields

Supported Operations

Mock Server

How It Works

Mock Data Generation

Dev Seeds Integration

Unsupported Operations

Best Practices

When to Use Connectors

Read-Only Pattern

Naming Conventions

Environment Variable Management

Mixing Cosmos DB and Connector Models

Limitations