Connect Your Work with Dify and MCP! A Guide to the Shortest Route for Turning Apps into MCP Servers and Integrating External MCPs
If you have needs like "I want to link email, databases, and business tools with generative AI" or "I want to try out the trending MCP," then Dify is the perfect fit.
Dify, which allows you to build full-scale generative AI apps without code, now supports MCP! You can turn the apps you create into MCP servers or integrate external MCP servers into your apps and more.
In this article, we will introduce how to use and apply Dify's MCP-related features.
This note is recommended for those who want to smoothly link various external tools with generative AI. By reading to the end, you will be able to understand the following!
1️⃣ How to turn your created apps into MCP servers
We will explain the method, including settings in Dify and operation checks, with a demonstration.
2️⃣ How to integrate external MCP servers into Dify apps
We will introduce two ways to implement MCP servers: the HTTP method and the SSE method.
3️⃣ Use cases for Dify x MCP
We will show you the optimal use cases, focusing on examples of external MCP servers and Dify apps.
📌 To those reading this Note
At WEEL, we hold various seminars for those who want to incorporate generative AI into their business and operations. For those wondering, "I want to know what kind of applications are possible" or "How should I start in my company?", we introduce
know-how from a field perspective and the latest case studies. ✅
Participate in the seminar
What is "Dify" in the first place?
"Dify" is a platform that allows you to build full-scale generative AI apps without code.
Its features are as follows, and the appeal lies in being able to create advanced chatbots and AI agents without programming.
Intuitive app construction using prompts and flows
Available in both cloud and local environments
Wide range of generative AI models can be integrated into apps
Various external services such as Slack and Gmail can be integrated into apps
With Dify, even non-engineer field staff can create their own generative AI apps. By skipping full-scratch development, you can automate miscellaneous non-core tasks.
ミニAIエージェントで事足りる、という場面がどんどん増えてきた。ChatGPT、Gemini、Notion、Dify。各ツールにAI機能が標準搭載されてきて、ノーコードでちょっとした効率化や自動化ができてしまう。フルスクラッチ開発が必要なのは、先進的な新規事業か、コア業務のがっつり効率化くらいになりそう。… pic.twitter.com/OA90Yq3M9n
— 田村洋樹 | WEEL(ウィール)代表 | 生成AIの企業事例を発信 (@tamtam_weel) January 26, 2026
In addition, if you want to know more about the pros and cons of Dify, please check the article below as well.
What is "MCP" that can also be used with Dify?
Dify also supports the trending "MCP (Model Context Protocol)".
MCP is a technology for integrating external tools such as web searches, email, and databases to work with generative AI.
Specifically, it is a mechanism that allows generative AI to use the functions of various external tools without the need for specially designed code or prompts.
For other detailed terms regarding MCP, please refer to the table below.
MCP, which has the above features, can also be used in Dify; you can publish the apps you create as MCP servers and also integrate external tool MCP servers into your apps.
How to turn a Dify app into an MCP server
From here, I will introduce how to actually use Dify to publish an app as an MCP server.
Publish via URL
To publish an app created in Dify as an MCP server, first click "Publish Update" in the "Publish" section from the app creation screen.
Next, click the publishing settings section (the red framed area in the figure below) on the left sidebar of the creation screen.
Then, the publishing settings panel will expand. From there, click the switch in the "MCP Server" section.
Next, the MCP server settings screen will pop up. To ensure the MCP client understands the functions of your Dify app, fill in the "Description" and "Parameters" fields here.
Finally, select "Enable MCP Server" to complete the process of turning your app into an MCP server.
From then on, you will be able to copy the server URL from the MCP server section of the publishing settings panel.
Note that the MCP Server URL published by Dify includes authentication information in advance, and this URL itself acts as an access key.
Checking the publishing status
Next, I will check whether the published MCP server is operating normally from Dify.
First, please open the publishing settings panel from the creation screen of the app you just turned into an MCP server. From there, click "API Key" in the "Backend Service API" section.
Next, select "+ Create New Secret Key".
Then, a new API secret key will be issued as shown below. Copy this to your clipboard.
Next, I will check the operating status of the MCP server using the API key I just obtained and the MCP server URL published earlier.
Now, let's go ahead and replace the following curl command with the API key and MCP server URL you obtained from Dify, and run it on the command line.
curl -X POST Dify App MCP Server URL \
-H "Content-Type: application/json" \
-H "Authorization: Bearer Issued API Key" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/list"
}'
The result of running the above command on the command line (Mac Terminal) will look like this.
{"jsonrpc": "2.0", "id": 1, "result": {"tools": [{"name": "Document OCR", "description": "Extracts the issuer name, date, product name, quantity, unit price, and total amount from uploaded document images (estimates/invoices).", "inputSchema": {"type": "object", "properties": {}, "required": []}}]}}
When we decode this...
array (
'jsonrpc' => '2.0',
'id' => 1,
'result' =>
array (
'tools' =>
array (
0 =>
array (
'name' => 'Form OCR',
'description' => 'Extracts the issuer name, date, product name, quantity, unit price, and total amount from uploaded form images (quotes/invoices).',
'inputSchema' =>
array (
'type' => 'object',
'properties' =>
array (
),
'required' =>
array (
),
),
),
),
),
)
As you can see, the MCP server description we set up earlier can be confirmed even from outside Dify.
All that's left is to integrate this MCP server into any generative AI tool of your choice!
How to use external tool MCP servers within Dify
In Dify, you can also use external MCP servers as tools in app types like "Agent" or in the "Agent" block of Chatflow / Workflow.
Next, let's take a look at how to introduce such MCP servers derived from external tools into Dify.
For the HTTP method
Dify comes with built-in functionality for introducing various MCP servers (MCP client functionality). Here, you can introduce simple HTTP-based MCP servers into Dify.
To use Dify's native MCP client functionality, first open the "Tools" screen and click on the "MCP" section. Then, select "Add MCP Server (HTTP)".
Next, the following pop-up screen will appear, so enter the various information for the MCP server you want to introduce into Dify (such as the server URL and token).
Finally, click "Add and Authorize" to complete the setup.
For the SSE method
In Dify, it is also possible to introduce SSE-based MCP servers using the "MCP SSE / StreamableHTTP"plugin.
To introduce an SSE-based MCP server, first open the "Plugins" screen and select the "Explore Marketplace" tab.
Next, search for and install the "MCP SSE / StreamableHTTP" plugin.
Then, open the Dify Tools screen and select the plugin you just installed (in this screen, it is MCP tools).
At the same time, click on "API Key Authentication Settings" displayed on the right side of the screen.
Finally, enter the information for the MCP server you want to introduce in JSON format from the pop-up screen and click "Save" to complete the introduction.
From then on, you will be able to use the introduced MCP server as a tool from the "Agent" app type or the "Agent" block in flow-based apps.
Use cases for Dify × MCP
Here, we will introducethe combined power of Dify and MCP,focusing on the example of 'Zapier (Zapier MCP)'and its use cases.
Multi-tool agent via Zapier integration
Dify is a perfect match for 'Zapier (Zapier MCP),' a tool that integrates multiple web services and connects them to generative AI. First, Zapier allows for centralized management of Slack, Gmail, Google Calendar, various databases, and more. Furthermore, you can turn the functions of each service into an MCP server and publish them via 'Zapier MCP.'
This Zapier MCP server is also compatible with Dify, and
you can call the functions of each service bundled by Zapier from Dify agents/agent blocks. The appeal here is that you can link multiple tools with Dify using just Zapier MCP, without needing to install plugins for each individual tool.
With such a Dify x Zapier MCP combination,
you can automate a wide range of tasks, from business communications on Slack to administrative work in spreadsheets and databases.
Database integration
With the Dify x Zapier MCP combination, it is also possible to link generative AI apps within Dify to databases and knowledge on external cloud services. This supports everything from structured data on Snowflake or AWS to documents on Google Drive.
In addition to being able to call up company-specific information just like RAG (Dify's knowledge feature),
you can easily perform information updates, which was difficult with RAG.
Calendar integration
By utilizing Zapier MCP, you can also link Dify with external calendars (such as Google Calendar). This allows you to
semi-automate tasks such as transcribing appointments received via email or Slack into your calendar, or conversely, sharing schedules from your calendar via email or Slack.
For those who want to know more about Dify use cases in general, please also check the article below.
Frequently Asked Questions (FAQ) about Dify & MCP
Finally, we will answer frequently asked questions (FAQ) about Dify and MCP.
Q. What is the pricing for Dify? Can I use it for free?
A. The web version of Dify is basically free to use, and the local version is completely free.
However, using AI models and external tools incurs separate costs such as API fees, and the web version has limitations based on the pricing plan regarding the number of times created apps can be used and the number of times external tools can be called.
The same applies to MCP; in the case of the web version of Dify, there are plan-specific limitations for both turning apps into MCP servers and using MCP servers within apps.
Q. What is the difference between MCP and API?
A. The differences between MCP and API are shown in the table below. By adopting MCP, linking generative AI with various APIs becomes easier.
Q. What is the difference between MCP and RAG?
A. Both MCP and RAG allow you to provide generative AI with information outside of its training scope. The differences between the two are shown in the table below.
For those who want to know more about the differences between Dify and other tools, please also check the article below.
Easier external service integration with Dify and MCP!
In this note, we introduced how to use MCP in Dify.
Dify supports MCP, allowing you to publish apps you've created as MCP servers or integrate external MCP servers into your apps.
In particular, the combination with Zapier MCP is powerful, enabling you to unify Dify's integration with multiple tools like Google Calendar and Slack.
Please consider this as an alternative to plugins!
Additionally, the WEEL website has a comprehensive article on Dify, so if you want to grasp the big picture, please take a look below.
At WEEL, we hold various seminars for those who want to incorporate generative AI into their work and business.
For those wondering, "I want to know how it can be used" or "How should we start in our company?", we introduce know-how and the latest case studies from a practical perspective.
If you want to be the first to know the latest news, case studies, and event information about generative AI, please subscribe to WEEL's free newsletter!
💡 For those worried about where to start with AI utilization
"How do you actually use AI agents?" "I want to try making a chatbot, but where should I start...?"
For those people, we offer free individual consultations! We will talk to you in an easy-to-understand way, incorporating real-world examples.