MCP Integration

MCP Integration

Model Context Protocol (MCP) enables AI assistants like Claude to interact with external tools and data sources. This guide covers setting up the Airbyte Agent MCP server for use with Claude Code and Claude Desktop.

Contents

• Overview
• MCP Server Location
• Available MCP Tools
• Configuration
• Claude Code Setup
• Claude Desktop Setup
• Running the Server Manually
• Example Usage with Claude
• Troubleshooting
• Development
• Related Documentation

Overview

The airbyte-agent-mcp server exposes Airbyte Agent Connectors as MCP tools, allowing Claude to:

• Discover available connectors
• Describe connector capabilities (entities, actions, parameters)
• Execute operations on any configured connector

MCP Server Location

The MCP server is included in this repository:

airbyte-agent-connectors/
└── airbyte-agent-mcp/
    ├── airbyte_agent_mcp/         # Server implementation
    ├── configured_connectors.yaml.example
    ├── pyproject.toml
    └── README.md

Available MCP Tools

execute

Execute an operation on a configured connector.

{
  "tool": "execute",
  "arguments": {
    "connector_id": "stripe",
    "entity": "customers",
    "action": "list",
    "params": {"limit": 10}
  }
}

Parameters:

• connector_id (required): ID of the configured connector
• entity (required): Entity to operate on (e.g., customers, issues)
• action (required): Action to perform (e.g., list, get, create)
• params (optional): Parameters for the operation

Returns: Operation result with data or error message.

list_entities

List all available entities for a connector.

{
  "tool": "list_entities",
  "arguments": {
    "connector_id": "github"
  }
}

Returns: Array of entity names with descriptions.

describe_entity

Get detailed schema for an entity including available actions and parameters.

{
  "tool": "describe_entity",
  "arguments": {
    "connector_id": "stripe",
    "entity": "customers"
  }
}

Returns: Entity schema with actions, parameters, and types.

validate_operation

Check if an operation is valid before executing.

{
  "tool": "validate_operation",
  "arguments": {
    "connector_id": "stripe",
    "entity": "customers",
    "action": "get",
    "params": {"id": "cus_xxx"}
  }
}

Returns: Validation result with any errors.

Configuration

Step 1: Create configured_connectors.yaml

Copy the example and customize:

cd airbyte-agent-mcp
cp configured_connectors.yaml.example configured_connectors.yaml

Edit configured_connectors.yaml:

connectors:
  # Stripe - API Key authentication
  - id: stripe
    type: local
    connector_name: stripe
    description: "Stripe payment processing"
    secrets:
      api_key: STRIPE_API_KEY

  # GitHub - Personal Access Token
  - id: github
    type: local
    connector_name: github
    description: "GitHub repositories and issues"
    secrets:
      token: GITHUB_ACCESS_TOKEN

  # Salesforce - OAuth
  - id: salesforce
    type: local
    connector_name: salesforce
    description: "Salesforce CRM"
    secrets:
      client_id: SALESFORCE_CLIENT_ID
      client_secret: SALESFORCE_CLIENT_SECRET
      refresh_token: SALESFORCE_REFRESH_TOKEN

  # HubSpot - Private App
  - id: hubspot
    type: local
    connector_name: hubspot
    description: "HubSpot CRM and marketing"
    secrets:
      access_token: HUBSPOT_ACCESS_TOKEN

  # Slack - Bot Token
  - id: slack
    type: local
    connector_name: slack
    description: "Slack messaging"
    secrets:
      token: SLACK_BOT_TOKEN

Step 2: Create .env File

Create .env in the airbyte-agent-mcp directory (no template provided, create from scratch):

# Stripe
STRIPE_API_KEY=sk_live_your_key_here

# GitHub
GITHUB_ACCESS_TOKEN=ghp_your_token_here

# Salesforce
SALESFORCE_CLIENT_ID=your_client_id
SALESFORCE_CLIENT_SECRET=your_client_secret
SALESFORCE_REFRESH_TOKEN=your_refresh_token

# HubSpot
HUBSPOT_ACCESS_TOKEN=pat-na1-xxx

# Slack
SLACK_BOT_TOKEN=xoxb-your-token

Configuration Options

Loading from Registry (Recommended)

- id: stripe
  type: local
  connector_name: stripe  # Loads from Airbyte registry
  secrets:
    api_key: STRIPE_API_KEY

Pinning to Specific Version

- id: stripe
  type: local
  connector_name: stripe
  version: 0.1.6  # Pin to specific version
  secrets:
    api_key: STRIPE_API_KEY

Loading from Local Path

- id: my_api
  type: local
  path: ./connectors/my-api/connector.yaml  # Local file path
  secrets:
    token: MY_API_KEY

Multi-Auth Connectors

For connectors with multiple auth methods, specify the scheme:

# Zendesk with OAuth
- id: zendesk-oauth
  type: local
  connector_name: zendesk-support
  auth_scheme: zendeskOAuth
  config_values:
    subdomain: yourcompany
  secrets:
    access_token: ZENDESK_ACCESS_TOKEN

# Zendesk with API Token
- id: zendesk-api
  type: local
  connector_name: zendesk-support
  auth_scheme: zendeskAPIToken
  config_values:
    subdomain: yourcompany
  secrets:
    email: ZENDESK_EMAIL
    api_token: ZENDESK_API_TOKEN

Claude Code Setup

Option 1: Project-Scoped (Recommended)

Add the MCP server to your project:

cd /path/to/your/project
claude mcp add airbyte-agent-mcp --scope project

When prompted, provide the server configuration:

{
  "command": "uv",
  "args": [
    "--directory",
    "/path/to/airbyte-agent-connectors/airbyte-agent-mcp",
    "run",
    "airbyte_agent_mcp"
  ]
}

Option 2: Global Configuration

Add to ~/.claude.json:

{
  "mcpServers": {
    "airbyte-agent-mcp": {
      "type": "stdio",
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/airbyte-agent-connectors/airbyte-agent-mcp",
        "run",
        "airbyte_agent_mcp"
      ],
      "env": {}
    }
  }
}

Option 3: With Custom Paths

If your config files are in different locations:

{
  "mcpServers": {
    "airbyte-agent-mcp": {
      "type": "stdio",
      "command": "python",
      "args": [
        "-m",
        "airbyte_agent_mcp",
        "/path/to/configured_connectors.yaml",
        "/path/to/.env"
      ],
      "cwd": "/path/to/airbyte-agent-connectors/airbyte-agent-mcp"
    }
  }
}

Claude Desktop Setup

Add to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "airbyte-agent-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/airbyte-agent-connectors/airbyte-agent-mcp",
        "run",
        "airbyte_agent_mcp"
      ]
    }
  }
}

Restart Claude Desktop after making changes.

Running the Server Manually

For testing or development:

cd airbyte-agent-connectors/airbyte-agent-mcp

# Install dependencies
uv sync

# Run the server
uv run airbyte_agent_mcp

With custom paths:

uv run python -m airbyte_agent_mcp /path/to/config.yaml /path/to/.env

Example Usage with Claude

Once configured, you can ask Claude to interact with your connectors. Here are example conversations showing how Claude uses the MCP tools.

Discovering Available Connectors

You: "What connectors are available?"

Claude: Uses list_entities tool to discover configured connectors

I can see you have the following connectors configured:

• stripe - Stripe payment processing
• github - GitHub repositories and issues
• hubspot - HubSpot CRM and marketing

Exploring Connector Capabilities

You: "What can I do with the Stripe connector?"

Claude: Uses describe_entity tool for each entity

The Stripe connector supports these entities and actions:

customers

• list - List all customers (params: limit, starting_after)
• get - Get customer by ID (params: id)
• create - Create new customer (params: email, name, metadata)
• api_search - Search customers (params: query)

invoices

• list - List invoices (params: limit, customer)
• get - Get invoice by ID (params: id)

Would you like me to perform any of these operations?

Executing Operations

You: "List my top 10 Stripe customers"

Claude: Uses execute tool

{
  "tool": "execute",
  "arguments": {
    "connector_id": "stripe",
    "entity": "customers",
    "action": "list",
    "params": {"limit": 10}
  }
}

Here are your top 10 Stripe customers:

1. Jane Smith (jane@example.com) - Created Jan 15, 2024
2. Acme Corp (billing@acme.com) - Created Jan 10, 2024 ...

Using Validation Before Execution

You: "Search for customers with email containing 'acme'"

Claude: Uses validate_operation first, then execute

{
  "tool": "validate_operation",
  "arguments": {
    "connector_id": "stripe",
    "entity": "customers",
    "action": "api_search",
    "params": {"query": "email~'acme'"}
  }
}

Validation passed. Executing search...

Multi-Connector Workflows

You: "Find GitHub issues labeled 'bug' and check if they have matching Jira tickets"

Claude: Uses execute on github connector, then jira connector

I found 5 GitHub issues labeled 'bug':

1. #123 - Login timeout issue
2. #124 - Dashboard not loading

...

Checking Jira for matching tickets...

• #123 matches PROJ-456 (In Progress)
• #124 has no matching Jira ticket

Common Prompts

Troubleshooting

Server Not Starting

1. Check Python/uv installation:

   uv --version
   python --version
   

2. Verify paths in configuration:

   • Ensure the --directory path points to airbyte-agent-mcp
   • Check that configured_connectors.yaml exists

3. Check for syntax errors:

   cd airbyte-agent-mcp
   uv run python -c "import yaml; yaml.safe_load(open('configured_connectors.yaml'))"
   

Connection Errors

1. Environment variables not loaded:

   • Ensure .env file exists in airbyte-agent-mcp directory
   • Check variable names match those in configured_connectors.yaml

2. Invalid credentials:

   • Verify API keys are valid and not expired
   • Check OAuth tokens haven't expired

Tool Execution Failures

1. Entity not found:

   • Use list_entities to see available entities
   • Check spelling and case sensitivity

2. Missing parameters:

   • Use describe_entity to see required parameters
   • Ensure all required fields are provided

Development

Running Tests

cd airbyte-agent-mcp
uv sync --all-extras
uv run pytest

Code Formatting

uv run ruff format .
uv run ruff check .

Adding a New Connector

1. Add connector to configured_connectors.yaml:

   - id: new_connector
     type: local
     connector_name: new-connector
     secrets:
       token: NEW_CONNECTOR_TOKEN
   

2. Add credentials to .env:

   NEW_CONNECTOR_TOKEN=your_token
   

3. Restart the MCP server

Related Documentation

• Getting Started - Install and configure connectors
• Entity-Action API - Understand the API patterns
• Authentication - Set up credentials
• Troubleshooting - Common issues and solutions
• airbyte-agent-mcp README - Server-specific documentation