Dify

What is Dify?

Dify is an agentic workflow builder to create, deploy, and manage autonomous AI agents and RAG (retrieval-augmented generation) pipelines. The product is positioned as a single place to build “production-ready” agentic workflows and related components.

Its core purpose is to help teams go from a workflow concept to something they can publish and run, while connecting models, data retrieval, and external tools into a cohesive application flow.

Key Features

• Drag-and-drop workflow creation: Build AI apps and workflows visually, including workflows designed to handle diverse tasks and evolving needs.
• Support for multiple global LLMs: Access, switch, and compare different large language models, including open-source and proprietary options.
• RAG pipeline building (“Get Your Data LLM Ready with RAG”): Prepare application data for LLM use by incorporating retrieval into the workflow.
• Integration via tools and plugins (“Add Wings with Tools”): Expand what an AI application can do by adding tools/plugins.
• Native MCP integration: Bridge external APIs, databases, and services using standardized MCP protocols, including support for HTTP-based MCP services (protocol dated 2025-03-26) and pre-authorized/auth-free modes.
• Publish workflows/agents as an MCP server (“Publish as an Universal MCP Server”): Expose a Dify-built workflow or agent so it can be accessed across unlimited MCP clients.
• Integrations and observability in one place: The site describes Dify as offering agentic workflows, RAG pipelines, integrations, and observability together.

How to Use Dify

1. Start building a workflow using the visual (drag-and-drop) builder to define the steps of your AI application.
2. Choose and configure LLMs you want the workflow to use, with the option to access and compare models.
3. Add RAG components to connect your data to the LLM portions of the workflow.
4. Attach tools/plugins and/or connect external services via MCP so the workflow can take actions or fetch information.
5. Publish the workflow using Dify’s available publishing options, including the option to publish as a universal MCP server for broader client access.

Use Cases

• Autonomous agent workflow for multi-step tasks: Create a workflow that chains multiple steps (reasoning, tool use, and actions) to handle tasks that require more than a single prompt.
• RAG-powered support or knowledge assistant: Build an application where retrieval from your data sources supports the generation performed by the LLM.
• Tool-augmented assistants: Expand an AI app beyond text generation by adding tools/plugins so the workflow can perform additional operations.
• Connecting business systems through MCP: Use native MCP integration to access external APIs, databases, and services using standardized MCP protocols.
• Making an internal workflow reusable across MCP clients: Publish a workflow/agent as a universal MCP server so other MCP clients can consume it.

FAQ

Is Dify limited to one kind of AI app (chat only)? No. The site describes building agentic workflows and RAG pipelines, not just chat interactions.

Can I use different LLM providers in the same workflow setup? The product is described as allowing access, switching, and comparison of different LLMs (including open-source and proprietary), suggesting model flexibility during workflow creation.

How does Dify connect my data to the LLM? Dify includes RAG capabilities (“Get Your Data LLM Ready with RAG”), indicating you can configure retrieval so the LLM can use your data in generation.

What is MCP integration used for in Dify? MCP integration is described as a way to bridge external APIs, databases, and services using standardized MCP protocols. It also supports publishing a workflow/agent as an MCP server.

Does Dify support HTTP-based MCP services? Yes. The page states support for HTTP-based MCP services with protocol 2025-03-26, including pre-authorized and auth-free modes.

Alternatives

• Low-code LLM workflow builders: Tools that provide visual builders for connecting prompts, models, and retrieval. These typically focus on app creation but may vary in how they support agent patterns and MCP-style server publishing.
• RAG-focused orchestration platforms: Solutions centered on building retrieval and document pipelines, often with less emphasis on multi-tool agent workflows or standardized server interfaces.
• API-first agent frameworks and SDKs: Developer-focused frameworks where you implement agent logic and integrations in code. These can offer greater control but require more engineering effort than a visual workflow builder.
• General automation platforms with AI add-ons: Workflow automation tools that can incorporate LLM steps and connectors. They may be broader for automation, but may not provide the same agentic workflow + MCP publishing orientation described for Dify.