fal MCP Server – Generative Media Model Design
fal's official MCP server gives AI assistants direct access to the fal platform for discovering and running generative-media models. Use it to search models, inspect schemas, upload inputs, run image or video jobs, manage long-running requests, check pricing, and build creative media pipelines.
Overview
fal provides an official hosted MCP server for connecting AI assistants to the fal generative-media platform. The server gives compatible clients access to fal's model catalog, documentation, schemas, pricing, file upload flow, and model execution APIs so an agent can search, choose, run, and chain media models from the conversation.
What the MCP server enables
The fal MCP server exposes nine documented tools across discovery, execution, and utility workflows:
- Search fal's catalog of 1,000+ models by keyword or category.
- Retrieve full input and output schemas for a selected model.
- Check model pricing before running a job.
- Search fal documentation for guides, examples, and API references.
- Run a model and wait for the result.
- Submit long-running jobs for video, 3D, or training workflows.
- Check job status, retrieve results, or cancel a running job.
- Upload a local file or URL to fal's CDN for model input.
- Recommend models from a natural-language task description.
The catalog covers many creative categories, including text-to-image, image-to-video, text-to-video, text-to-speech, image-to-3D, image editing, upscaling, and other generative-media workflows.
When to use it
Use fal MCP when a design or content workflow needs flexible access to multiple generative-media models rather than a single fixed image API. Typical examples include generating campaign visuals, turning an image into a short video, comparing model outputs, checking video-generation cost before running, uploading product photos for editing, selecting the best background-removal model, or chaining generation, upscaling, and analysis steps.
Connection and authentication
The official Streamable HTTP endpoint is:
https://mcp.fal.ai/mcp
Every request uses the user's own fal API key through the Authorization header:
Authorization: Bearer YOUR_FAL_KEY
fal states that the hosted server is stateless, sends the API key per request, and does not store the key. OAuth authentication for Claude Desktop and claude.ai Custom Connectors is not yet supported, so use clients that can provide custom headers with Streamable HTTP, such as Claude Code, Cursor, or Windsurf.
Key considerations
The MCP server itself is free, but model runs use standard fal pricing. Long-running jobs should use submit_job and check_job to avoid timeouts. Access is limited to what the fal public API allows with the supplied key. Store the API key securely, avoid sharing MCP client configuration files, and review generated media for quality, rights, consent, brand fit, safety, and policy compliance before publication.
Supported Transports
streamable_http
URL: https://mcp.fal.ai/mcp
Frequently Asked Questions
- When should an AI agent use the fal MCP server?
- Use it when a workflow needs to discover, compare, price, run, or chain fal-hosted generative-media models for image generation, image editing, video generation, audio, 3D, upscaling, or related creative tasks.
- What does the fal MCP server add to an AI agent's capabilities?
- It gives the agent live access to fal's model catalog, model schemas, pricing information, documentation search, file upload, model execution, queued jobs, result retrieval, cancellation, and model recommendations.
- What can an AI agent access or manage through fal MCP?
- The agent can search more than 1,000 fal models, inspect model parameters, upload files to fal's CDN, run models, submit long-running jobs, check or cancel jobs, fetch results, search docs, and recommend models for a requested task.
- How is authentication configured for the fal MCP server?
- The hosted server requires a fal API key sent on every request as Authorization: Bearer ${FAL_KEY}. fal documents that the stateless hosted server does not store the key and only calls the public fal API on the user's behalf.
- Which transport should be used for the fal MCP server?
- Use Streamable HTTP with https://mcp.fal.ai/mcp and an Authorization bearer header. OAuth for Claude Desktop and claude.ai Custom Connectors is not yet supported, so use clients that allow custom headers.