Connector Framework¶

Connectors are the bridges between Manifest Platform and external systems. They provide a standardized interface for reading from and writing to databases, SaaS APIs, file storage, messaging systems, and analytics platforms -- all with managed credentials, rate limiting, and observability.

What Connectors Do¶

A connector wraps an external system's API behind a consistent abstraction:

graph LR
    Agent["Agent"] --> Tool["MCP Tool"]
    Workflow["Workflow"] --> Op["Operation Call"]
    Dataset["Dataset Source"] --> Op

    Tool --> Instance["Connector Instance<br/>(config + auth profile)"]
    Op --> Instance

    Instance --> Creds["Credentials<br/>(org / workspace / user / agent)"]
    Instance --> Connector["Connector Descriptor<br/>(operations + auth + schema)"]

    Connector --> API["External API<br/>(Jira, Snowflake, S3, etc.)"]

Instead of each agent or workflow implementing its own HTTP calls, authentication, and error handling, you define a connector once and reuse it everywhere.

Key Concepts¶

Connector Descriptor¶

The descriptor is the blueprint for a connector. It defines the external system's API operations, authentication methods, rate limits, and capabilities in a declarative YAML or JSON document. Descriptors are validated by the platform and versioned in the connector catalog.

Connector Instance¶

An instance is a configured deployment of a connector descriptor. It binds the descriptor to specific credentials and configuration for a particular environment. One connector descriptor (e.g., "Jira Cloud") can have multiple instances (e.g., "Jira - Engineering", "Jira - Support").

Operations¶

Operations are the individual API endpoints a connector exposes. Each operation has a unique ID, HTTP method, endpoint path, parameters, and authentication reference. Operations are the atomic units that agents and workflows invoke.

Credentials¶

Credentials are encrypted authentication tokens stored per-instance. The platform supports API keys, basic auth, OAuth2 (client credentials and authorization code), and service accounts. Credentials are never returned in API responses -- only metadata is visible.

Connector Categories¶

Category	Examples	Use Cases
Database	PostgreSQL, MySQL, Snowflake, BigQuery	Query tables, run SQL, sync data
SaaS	Jira, Salesforce, HubSpot, Slack	CRUD operations, search, webhooks
File Storage	S3, GCS, Azure Blob, SFTP	Upload, download, list files
Messaging	Kafka, RabbitMQ, SQS	Publish and consume messages
Analytics	Mixpanel, Amplitude, Google Analytics	Pull metrics, export reports
Custom	Internal APIs, proprietary systems	Any HTTP-based integration

Connector Capabilities¶

Each connector declares its capabilities in a capability matrix:

Capability	Description
Sync modes	`full_refresh`, `incremental`, `streaming`
Schema discovery	Automatically detect table/field schemas
Incremental bookmark	Track sync position for incremental loads
Change data capture	Stream real-time changes from source
Custom SQL	Execute arbitrary SQL queries
File ingestion	Read and parse file uploads
Max parallel tasks	Concurrent operation limit

Authentication Types¶

Connectors support six authentication methods:

Auth Type	Description	Example
`none`	No authentication required	Public APIs
`api_key`	Static API key in header or query	Stripe, SendGrid
`basic`	Username and password (HTTP Basic)	JIRA Server
`oauth2_client_credentials`	Machine-to-machine OAuth2	Google Cloud APIs
`oauth2_authorization_code`	User-delegated OAuth2 with consent flow	Salesforce, HubSpot
`service_account`	Service account credentials (e.g., JSON key file)	GCP, Firebase

Credential chains

Some systems require multi-step authentication (e.g., exchange a username/password for a token, then exchange that token for a bearer token). Connectors support credential chains that automate these flows with caching and TTL management.

How Connectors Are Used¶

By Agents (via MCP)¶

Connectors with MCP enabled generate tools that agents can call directly. The agent says "I need to search Jira tickets" and the platform translates that into a connector operation call with proper authentication.

By Workflows¶

Workflow nodes can invoke connector operations as steps in an automation pipeline. A workflow might pull data from a database connector, transform it, and push it to a SaaS connector.

By Datasets¶

Datasets can be backed by connector instances as data sources. The dataset query API routes through the connector to fetch live data with optional caching and field mapping.

Next Steps¶

Building Connectors

Define connector descriptors with operations, auth, and schemas.

Build a connector
Instances & Credentials

Provision instances, manage credentials, and test connections.

Manage instances
MCP Configuration

Expose connector operations as agent tools via MCP.

Configure MCP
Testing

Test connections, operations, and debug issues.

Test connectors