Skip to main content
Fal is an AI inference platform that hosts hundreds of generative AI models. The Fal MCP server lets you search for models, inspect their schemas, submit inference jobs, and retrieve results across image, video, audio, speech, and 3D generation categories.

​
What Can It Do?

  • Search AI models by category (text-to-image, text-to-video, text-to-audio, text-to-speech, text-to-3D)
  • Inspect model schemas to understand accepted input parameters before running a model
  • Submit inference requests that queue on fal.ai and return a request ID for polling
  • Poll for results including generated images, videos, audio files, and 3D assets with download URLs

​
Where to Use It

Add Fal as a tool to any agent. The agent can search for the right model, check its schema, submit a generation request, and poll for results conversationally. To add an MCP tool to your agent:
  1. Open your agent’s configuration
  2. Click Add tools → Connect an app with MCP
  3. Search for the integration and select it
  4. Authenticate with your Fal account
You can control which tools your agent has access to. After adding an integration, click on it to enable or disable specific tools based on what your agent needs.

​
In Workflows (Via Agent Node)

For automated pipelines, use an Agent Node with Fal tools. This gives you the flexibility of an agent within a deterministic workflow.

​
As a Custom MCP Node

You can also create a standalone MCP node for a specific action. This generates a reusable node that performs one task, useful when you need the same operation repeatedly in workflows.
To create a custom MCP node:
  1. Go to your node library and search for the integration
  2. Click Create a node with AI
  3. Describe the specific action you want (e.g., “Generate an image using fal-ai/flux/dev”)
  4. Test the node and save it for reuse
Custom MCP nodes are single-purpose by design. For tasks that require multiple steps or dynamic decision-making, use an agent instead.

​
Available Tools

​
Run Model Credit Costs

The run_model tool cost depends on the model category:

​
How It Works — Asynchronous Generation

Fal does not return results instantly. It uses an asynchronous queue system, which means generation happens in the background and you retrieve the output separately once it’s ready. Here’s the full flow:
  1. Search for a model using search_models with a category like text_to_image
  2. Inspect the model’s accepted parameters using get_model_schema
  3. Submit a request using run_model — this queues the job on fal.ai and immediately returns a request_id. The generation has started, but the result is not available yet.
  4. Wait and poll using get_result with the request_id. The status will progress through IN_QUEUE → IN_PROGRESS → COMPLETED. You need to keep polling until the status reaches COMPLETED.
  5. Retrieve the output — once COMPLETED, the response contains download URLs for the generated content (images, videos, audio, 3D assets).
Generation is not real-time. After submitting a request, there is a waiting period while fal.ai processes the job. Image generation typically takes a few seconds, but video and 3D generation can take several minutes. The agent will submit the job, then poll periodically until the result is ready.
The download URLs returned in the result are temporary. Make sure to download or use the generated files promptly after retrieval.

​
Example Prompts

Use these with your agent or in the Agent Node: Create a product hero image:
Generate a social media video:
Create a voiceover for a demo:
Generate background music:
Create a 3D asset:
Batch-generate marketing visuals:

​
Troubleshooting

Agents handle the full async workflow automatically. When you ask “Generate a product photo of sneakers,” the agent will search for an image model, check its schema, submit the request, wait for it to finish processing, and then return the download URL — all without you needing to manage the polling yourself. If results seem off, check the agent’s step-by-step reasoning.

​
Need Help?

Use this integration directly in Claude or Cursor. Connect remotely via the Fal MCP server using credentials from your Connectors page.