Airbyte Agent Connectors

First, determine which mode the user needs:

Platform Mode (Airbyte Cloud)

Use when:

• Environment has AIRBYTE_CLIENT_ID + AIRBYTE_CLIENT_SECRET
• User wants connectors visible in the Airbyte UI at app.airbyte.ai
• User needs managed credential storage, context store, or multi-tenant deployments

See Agent Engine Platform docs and the hosted quickstart tutorial for full details.

OSS Mode (Open Source / Local SDK)

Use when:

• User wants to run connectors directly without platform integration
• User is doing quick development or prototyping
• User wants Claude Code/Desktop integration via MCP only

Ask if unclear: "Are you using Airbyte Platform (app.airbyte.ai) or open source connectors?"

Supported Connectors

51 connectors available. All connectors follow the same entity-action pattern: connector.execute(entity, action, params)

Full table: See references/connector-index.md for all 51 connectors with auth types, key entities, and documentation status.

If the connector is NOT in the index: Inform the user that this connector isn't available yet. Point them to GitHub issues.

Platform Mode Quick Start

For users with Airbyte Platform credentials.

Prerequisites

Get credentials from app.airbyte.ai > Settings > API Keys:

• AIRBYTE_CLIENT_ID
• AIRBYTE_CLIENT_SECRET

Create a Connector

from airbyte_agent_stripe import StripeConnector
from airbyte_agent_stripe.models import StripeAuthConfig

connector = await StripeConnector.create_hosted(
    external_user_id="user_123",
    airbyte_client_id="...",
    airbyte_client_secret="...",
    auth_config=StripeAuthConfig(api_key="sk_live_...")
)

Use Existing Connector

connector = StripeConnector(
    external_user_id="user_123",
    airbyte_client_id="...",
    airbyte_client_secret="...",
)
result = await connector.execute("customers", "list", {"limit": 10})

OSS Mode Quick Start

Install

# Using uv (recommended)
uv add airbyte-agent-github

# Or using pip in a virtual environment
python3 -m venv .venv && source .venv/bin/activate
pip install airbyte-agent-github

Use Directly

from airbyte_agent_github import GithubConnector
from airbyte_agent_github.models import GithubPersonalAccessTokenAuthConfig

connector = GithubConnector(
    auth_config=GithubPersonalAccessTokenAuthConfig(token="ghp_your_token")
)

result = await connector.execute("issues", "list", {
    "owner": "airbytehq",
    "repo": "airbyte",
    "states": ["OPEN"],
    "per_page": 10
})

Add to Claude via MCP

claude mcp add airbyte-agent-mcp --scope project

Entity-Action API Pattern

All connectors use the same interface:

result = await connector.execute(entity, action, params)
# For list actions: returns an envelope with result.data (list) and result.meta (pagination dict)
# For get/create/update actions: returns a raw dict (use dict access like result['name'])
# Raises exceptions on failure (RuntimeError, TypeError, ValueError, etc.)

Important: Error handling in agent tools. When registering execute() as a tool for an AI agent (PydanticAI, LangChain, etc.), always wrap the call in try/except and return the error message as a string. This prevents unhandled exceptions from crashing the agent and lets the LLM recover gracefully:

@agent.tool_plain
async def execute(entity: str, action: str, params: dict | None = None) -> str:
    try:
        return await connector.execute(entity, action, params or {})
    except Exception as e:
        return f"Error: {type(e).__name__}: {e}"

tool_utils and enable_hosted_mode_features: The @Connector.tool_utils decorator auto-generates the tool's docstring from the connector schema, including available entities, actions, parameters, and usage guidelines. In Platform Mode (default), the docstring includes search actions that query the context store and guidance to prefer cached search over direct API calls. In OSS Mode, pass enable_hosted_mode_features=False to exclude these — the context store is not available locally, so advertising search actions would cause errors:

# OSS Mode — excludes context store search actions from the tool docstring
@Connector.tool_utils(enable_hosted_mode_features=False)

# Platform Mode (default) — includes search actions and context store guidance
@Connector.tool_utils

Actions

Quick Examples

# List
await connector.execute("customers", "list", {"limit": 10})

# Get
await connector.execute("customers", "get", {"id": "cus_xxx"})

# Search
await connector.execute("repositories", "api_search", {
    "query": "language:python stars:>1000"
})

Pagination

async def fetch_all(connector, entity, params=None):
    all_records = []
    cursor = None
    params = params or {}

    while True:
        if cursor:
            params["after"] = cursor
        result = await connector.execute(entity, "list", params)
        all_records.extend(result.data)

        if result.meta and hasattr(result.meta, 'pagination'):
            cursor = getattr(result.meta.pagination, 'cursor', None)
            if not cursor:
                break
        else:
            break

    return all_records

Authentication Quick Reference

API Key Connectors

# Stripe
from airbyte_agent_stripe.models import StripeAuthConfig
auth_config=StripeAuthConfig(api_key="sk_live_...")

# Gong
from airbyte_agent_gong.models import GongAccessKeyAuthenticationAuthConfig
auth_config=GongAccessKeyAuthenticationAuthConfig(
    access_key="...", access_key_secret="..."
)

# HubSpot (Private App)
from airbyte_agent_hubspot.models import HubspotPrivateAppAuthConfig
auth_config=HubspotPrivateAppAuthConfig(private_app_token="pat-na1-...")

# HubSpot (OAuth)
from airbyte_agent_hubspot.models import HubspotOauth2AuthConfig
auth_config=HubspotOauth2AuthConfig(
    client_id="...", client_secret="...", refresh_token="..."
)

Personal Access Token

# GitHub
from airbyte_agent_github.models import GithubPersonalAccessTokenAuthConfig
auth_config=GithubPersonalAccessTokenAuthConfig(token="ghp_...")

# Slack
from airbyte_agent_slack.models import SlackAuthConfig
auth_config=SlackAuthConfig(token="xoxb-...")

OAuth (requires refresh token)

# Salesforce
from airbyte_agent_salesforce.models import SalesforceAuthConfig
auth_config=SalesforceAuthConfig(
    client_id="...", client_secret="...", refresh_token="..."
)

Per-connector auth details: Each connector reference in references/connectors/ includes the specific auth class name and fields.

Security Best Practices

• Never hard-code credentials in source files. Use environment variables or .env files.
• Use .env pattern: from dotenv import load_dotenv; load_dotenv()
• Rotate tokens regularly. Use short-lived tokens where possible.
• For production, use Platform Mode for managed credential storage.

Per-Connector Reference Documentation

• Connector Index -- full table of all connectors
• Per-connector references: references/connectors/

Each per-connector reference includes: package name and version, authentication details, available entities and actions, quick-start code snippets, and links to full GitHub documentation.

Reference Documentation

External Documentation

How References Are Generated

Reference docs are auto-generated by scripts/generate_skill_references.py. Run python scripts/generate_skill_references.py --all to regenerate all references, or --connector <name> for a specific connector.

Skill Metadata

• Author: Airbyte
• Version: 1.2.0
• License: Elastic-2.0
• Compatibility: Requires Python 3.11+. Recommends uv for package management.
• Repository: https://github.com/airbytehq/airbyte-agent-connectors
• MCP Server: airbyte-agent-mcp

Support

• Slack Community: slack.airbyte.com
• GitHub Issues: airbytehq/airbyte-agent-connectors
• Documentation: docs.airbyte.com/ai-agents