# Integration Hub Connect AgenticFlow to 300+ tools and services through secure connections. No coding required. ## Understanding Connections **Connections** are the foundation of AgenticFlow's integration system. A connection represents an authenticated link between AgenticFlow and an external service, enabling your agents and workflows to interact with that service. ### What Can Connections Do? Once configured, connections allow your agents and workflows to: * **Access external tools and services** - Connect to APIs, databases, and platforms * **Read and write data** - Retrieve and update information from connected services * **Execute actions** - Perform operations in third-party systems * **Retrieve information** - Fetch data on-demand during agent conversations * **Automate cross-platform workflows** - Chain actions across multiple services *** ## Two Types of Connections AgenticFlow supports two distinct types of connections, each designed for different integration scenarios: ### 1. MCP Connections **MCP (Model Context Protocol) Connections** connect to MCP servers - standardized integration servers that expose tools, prompts, and resources through a universal protocol. **Key Characteristics:** * **Connect once, get many tools** - A single MCP connection can provide dozens of tools * **Automatic tool discovery** - MCP servers expose their capabilities automatically; no manual configuration * **Standardized protocol** - All MCP servers follow the same interface, making integration consistent * **300+ services available** - One protocol connects to hundreds of services (Slack, GitHub, Google Drive, etc.) * **OAuth 2.0 support** - Secure authentication with automatic endpoint discovery * **Flexible deployment** - Connect to cloud-hosted or self-hosted MCP servers * **Dynamic capabilities** - MCP servers can provide tools, prompts, and context resources **How MCP Connections Work:** 1. You provide an MCP server URL (e.g., `https://mcp.slack.com/sse`) 2. AgenticFlow connects to the server and discovers all available tools 3. Tools from that MCP connection become available in your agents and workflows 4. The MCP server handles all API interactions with the third-party service **Example:** When you create an MCP connection to a Slack MCP server, it automatically exposes tools like "send message", "list channels", "get user info", etc. You don't configure individual tools - the MCP server provides them all. **MCP connections are the recommended integration method** due to their flexibility, extensive ecosystem, and simplified setup. 📖 **Learn more:** [What are MCPs?](https://docs.agenticflow.ai/integrations/overview) | [Browse MCP Integrations](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/README.md) *** ### 2. Direct API Connections **Direct API Connections** are built-in integrations for specific AI models and specialized services that require direct API access. **Key Characteristics:** * **Purpose-built for AI services** - Optimized for AI model providers and specialized tools * **Simple credential setup** - Just provide an API key or complete OAuth flow * **Direct API access** - AgenticFlow communicates directly with the service's API * **No MCP server required** - Native integration without intermediary servers **Available Direct API Connections:** * **AI Models** - OpenAI, Claude, Google Gemini, DeepSeek, Groq, Perplexity * **AI Services** - Replicate, FAL.ai, Straico, PixelML * **Specialized Tools** - Firecrawl (web scraping), Tavily (search), Telegram (messaging) **Authentication Methods:** * **API Keys** - Most AI services use simple API key authentication * **OAuth 2.0** - Services like YouTube use OAuth for secure delegated access **When to Use Direct API Connections:** * When integrating with AI model providers (OpenAI, Claude, etc.) * When using specialized services that require direct API access * When the service is not available through MCP 📖 **Learn more:** [Connection Providers](https://docs.agenticflow.ai/integrations/connection-providers) *** ## Quick Start ### [Getting Started with MCP Integrations](https://docs.agenticflow.ai/integrations/getting-started) Connect your first MCP tool in 5 minutes. ### [What are MCPs?](https://docs.agenticflow.ai/integrations/overview) Understanding the Model Context Protocol and how MCP integrations work. ### [Connection Providers](https://docs.agenticflow.ai/integrations/connection-providers) Complete reference of all connection providers with configuration details and credential links. *** ## Popular MCP Integrations ### [Browse All MCP Integrations](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/README.md) The following integrations are available through the Model Context Protocol (MCP). Each integration in the `mcp/` directory provides tools and capabilities that can be connected to your agents and workflows. **Communication**: * [Slack](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/slack.md) - Team messaging and notifications * [Gmail](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/gmail.md) - Email automation * [Discord](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/discord.md) - Community management * [Microsoft Teams](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/microsoft-teams.md) - Enterprise communication **Data & Storage**: * [Google Drive](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/google-drive.md) - File storage and sharing * [Dropbox](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/dropbox.md) - Cloud storage * [Google Sheets](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/google-sheets.md) - Spreadsheet automation * [Airtable](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/airtable.md) - Database management **CRM & Sales**: * [Salesforce](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/salesforce.md) - CRM automation * [HubSpot](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/hubspot.md) - Marketing and sales * [Pipedrive](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/pipedrive.md) - Sales pipeline * [Intercom](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/intercom.md) - Customer messaging **Productivity**: * [Notion](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/notion.md) - Workspace organization * [Trello](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/trello.md) - Project management * [Asana](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/asana.md) - Task management **Development**: * [GitHub](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/github.md) - Code repository * [GitLab](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/gitlab.md) - DevOps platform * [Jira](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/jira.md) - Issue tracking * [Linear](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/linear.md) - Project planning ### Additional Categories **Finance & Accounting** Invoicing, payments, accounting, expenses - browse [all MCP integrations](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/README.md) for financial tools. **Marketing & Analytics** Marketing automation, analytics, SEO tools - browse [all MCP integrations](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/README.md) for marketing platforms. **AI & Machine Learning** AI services, ML platforms, data science tools - browse [all MCP integrations](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/mcp/all/README.md) for AI/ML services. *** ## AgenticFlow as an MCP ### [Using AgenticFlow as an Integration](https://docs.agenticflow.ai/integrations/agenticflow-mcp) Use AgenticFlow agents and workflows from other applications: * MCP server setup * Exposing agents as tools * Workflow execution via MCP * Integration examples *** ## How Connections Work ### Creating a Connection All connections in AgenticFlow are **project-scoped**, meaning each connection belongs to a specific project and can only be used within that project. **For MCP Connections:** 1. Navigate to **Connections** in your project 2. Click **New Connection** → **MCP Connection** 3. Enter the **MCP server URL** (e.g., `https://mcp.example.com/sse`) 4. Select **transport type** (SSE or Streamable HTTP) 5. Choose **authentication method**: * **None** - For public MCP servers * **OAuth** - For secured servers (auto-discovers OAuth configuration) 6. Click **Validate & Create** - AgenticFlow tests the connection and discovers available tools 7. Connection is ready to use in agents and workflows **For Direct API Connections:** 1. Navigate to **Connections** in your project 2. Click **New Connection** → Select service (e.g., OpenAI, Claude, etc.) 3. Enter required credentials: * **API Key** - For most AI services * **OAuth** - For services like YouTube (follows OAuth flow) 4. Connection is validated and ready to use ### Using Connections in Agents 1. Open agent configuration → **Tools** tab 2. Add tools from your MCP connections or enable built-in tools 3. Agent automatically uses these tools during conversations 4. Tools appear in the agent's available capabilities 5. Results are returned to the agent context for reasoning ### Using Connections in Workflows 1. Add workflow nodes that require connections (e.g., AI model nodes, MCP tool nodes) 2. Each node requires a **specific connection type**: * **AI Model nodes** (OpenAI, Claude, etc.) → Require **Direct API Connections** * **MCP Tool nodes** → Require **MCP Connections** 3. Select the appropriate connection from the node's dropdown 4. Connection is validated against the node's required type before execution 5. Workflow uses the connection when executing the node 6. Outputs can be passed to subsequent nodes ### Using Connections in Workforce 1. Assign MCP tools to specific agents in your team 2. Each agent can use tools independently during task execution 3. Agents share results and coordinate tool usage 4. Connection is shared across all team members in the project *** ## Authentication Methods Connections use secure authentication to protect your credentials: ### **API Keys (Secret Texts)** * Simple key-based authentication * **Encrypted at rest** using AES-256 encryption * Stored securely per project * Used for: OpenAI, Claude, Replicate, Firecrawl, Tavily, and most AI services ### **OAuth 2.0** * Secure delegated access without sharing passwords * **Automatic token refresh** - Expired tokens are refreshed automatically * **Auto-discovery** - For MCP connections, OAuth endpoints are discovered automatically * **PKCE flow support** - Enhanced security for public clients * Used for: YouTube, MCP servers with OAuth ### **MCP OAuth Auto-Discovery** When creating an MCP connection with OAuth: 1. AgenticFlow queries the MCP server for OAuth metadata 2. Discovers authorization and token endpoints automatically 3. Registers AgenticFlow as an OAuth client with the server 4. Generates secure authorization URL with PKCE 5. Handles token exchange and refresh automatically 6. Validates connection and fetches available tools ### **No Authentication** * For public APIs and open MCP servers * No credentials required * Connection validated by testing server availability *** ## Connection Security AgenticFlow implements enterprise-grade security for all connections: ### **Encryption** * All credentials are **encrypted at rest** using AES-256 encryption * Encryption keys are managed securely and rotated regularly * OAuth tokens are encrypted before storage * Connections are decrypted only during execution ### **Project Isolation** * Connections are **scoped to projects** - cannot be accessed across projects * Each project maintains its own set of connections * Deleting a connection removes it from all agents and workflows in the project ### **Validation** * All connections are **validated on creation** - AgenticFlow tests connectivity * MCP connections verify server availability and fetch available tools * Direct API connections validate credentials before saving * Nodes validate connection category matches before execution ### **Best Practices** * **Use OAuth when available** - More secure than API keys * **Rotate API keys regularly** - Update credentials periodically * **Minimize permissions** - Request only necessary scopes for OAuth * **Audit connection usage** - Review which agents/workflows use each connection * **Delete unused connections** - Remove connections no longer needed *** ## Integration Examples ### Example 1: Slack Notification ``` Agent conversation → Trigger action → Slack MCP → Post message ``` ### Example 2: Google Sheets Data Lookup ``` User query → Agent → Google Sheets MCP → Retrieve data → Agent response ``` ### Example 3: CRM Update Workflow ``` Form input → Salesforce MCP → Create/update record → Email notification ``` ### Example 4: Multi-Tool Research ``` Agent → [Google Search, Wikipedia, Company Database] → Synthesize → Response ``` *** ## Troubleshooting Common integration issues: * **Authentication failures**: Verify credentials and permissions * **Rate limiting**: Implement backoff and retry logic * **Timeout errors**: Increase timeout settings or optimize requests * **Data format issues**: Check API documentation for expected formats * **Permission errors**: Review OAuth scopes and API permissions See [Troubleshooting Guide](https://docs.agenticflow.ai/support/troubleshooting) for more help. *** ## Featured Integration Use Cases ### Customer Service * **Slack + Zendesk + CRM**: Unified support experience * **Gmail + Calendar + Notion**: Automated email management ### Sales & Marketing * **HubSpot + LinkedIn + Gmail**: Automated outreach * **Salesforce + Google Sheets + Slack**: Sales reporting ### Development * **GitHub + Jira + Slack**: Development workflow automation * **GitLab + Sentry + PagerDuty**: Incident management ### Data & Analytics * **Google Analytics + Data Studio + Sheets**: Automated reporting * **Airtable + Zapier + Slack**: Data pipeline automation See [Use Cases](https://docs.agenticflow.ai/use-cases/10-use-cases) for more examples. *** ## Integration Limits ### Platform Limits * **Free Plan**: 10 integrations * **Pro Plan**: 100 integrations * **Team Plan**: 500 integrations * **Enterprise Plan**: Unlimited ### Rate Limits * Varies by integration * Respects third-party API limits * Automatic rate limiting protection * Enterprise plans: Higher limits See [Platform Limits](https://docs.agenticflow.ai/integrations/broken-reference) for details. *** ## Related Documentation * [Agents](https://docs.agenticflow.ai/ai-agents/03-agents) - Using tools in agents * [Workflows](https://docs.agenticflow.ai/workflows/04-workflows) - Integration nodes * [Workforce](https://docs.agenticflow.ai/workforce/05-workforce) - Tools in multi-agent teams * [Enterprise](https://docs.agenticflow.ai/integrations/broken-reference) - Enterprise integration management * [API Reference](https://docs.agenticflow.ai/developers/api) - Integration API *** ## Learn More * [AgenticFlow 101](https://docs.agenticflow.ai/learn/courses/agenticflow-101) - Foundation course * [Video Tutorials](https://github.com/PixelML/agenticflow-docs/blob/main/docs/02-learn/video-tutorials/README.md) - Integration walkthroughs * [Community](https://docs.agenticflow.ai/support/community) - Ask about integrations *** **Ready to connect your first tool?** [Start the Integration Guide →](https://github.com/PixelML/agenticflow-docs/blob/main/docs/07-integrations/getting-started.md)