An API lets one program request data or an action. A webhook tells another program when an event occurs. MCP can make capabilities discoverable to a compatible AI application. A single enquiry workflow may use all three, but a small project often needs only one or two.
Step-by-step setup
Connect an AI assistant to a Figma practice file
What you need A Figma account with access to a practice file, and a supported MCP host. The example uses the Codex app.
- Frame link
- AI app
- Figma MCP server
- Frame details
Prepare a test design
Open or create a practice file with one simple frame and give it a distinctive name, such as Practice Course Card.
Connect the service
In the Codex app, open Plugins and install Figma. Follow the sign-in and authorize it in the browser. The official remote MCP server is at
https://mcp.figma.com/mcp. It does not require the Figma desktop app.Point the assistant at one frame
Copy a link to the chosen frame. Paste it into the prompt below and ask for a read-only summary.
Check the result against the design
Compare the reported frame name and layout with Figma. Check that the assistant used a Figma tool. If it cannot open the link, confirm you can open the file with the account you authorized.
Copy this example
Use Figma MCP to inspect this practice frame: [paste Figma frame link].
Return the frame name and summarize its layout. Do not modify the file.
How to tell it worked
The host reports the right frame and its visible elements. You connected to a remote service here. Blender uses a local add-on, but both integrations give the AI app specific tools it can call.
Official setup references
Use an API when the caller knows what to request
Suppose your workflow needs an approved syllabus whenever it receives a course ID. If it already knows how to request that syllabus from the catalogue service, a direct API call can be sufficient. Validate the response and define what to do when the ID is missing.
Use a webhook to react to an event
An enquiry form can notify the workflow when a submission arrives, instead of making the workflow repeatedly check for new records. Use signed or authenticated events when the source supports them. Keep an event ID so replaying one submission does not create another identical task.
Use MCP when an AI app needs discoverable tools
An MCP server can describe catalogue tools and their expected inputs to an AI host. The host can then use those capabilities when suitable. The server may still call an ordinary catalogue API internally. The two interfaces solve different parts of the integration.
Combine them with clear boundaries
A webhook can start an enquiry workflow. A direct API can retrieve the record the workflow already knows it needs. If the agent must choose between syllabus and course-format lookups, those can be offered through MCP. Keep the action that changes records or sends a message under workflow control.
Try it yourself
Draw a form event, a known catalogue request and an optional AI tool lookup as separate boxes. Write down what starts each interaction and which program is responsible for checking the result.
Common questions
Can MCP replace an existing API?
Usually the API continues to provide the underlying application functions. An MCP server can present a selected set of those functions to an AI host.
Does a webhook answer a question?
A webhook primarily delivers an event. A later API call may be needed to retrieve the details or act on the event.
