Skip to main content
Every part of fal’s catalog and documentation is available as plain text, so an AI agent can read it directly instead of scraping HTML. Each surface is a different granularity or format — pick the smallest one that answers the question. All of them are public: no API key, no authentication header.
If your assistant supports the Model Context Protocol, Run MCP is usually a better fit than fetching these files: it can search models and check schemas on demand rather than loading a whole document.

​
The Surfaces

​
Per-Model Schemas

The highest-value surface for an agent mid-task is the per-model file. Append /llms.txt to the model page URL, fal.ai/models/<endpoint-id> — not to a /playground or /api variant of it:
It returns that endpoint’s current input and output schema with types, required and optional fields, defaults, value constraints, pricing, and generated snippets for cURL, Python, and JavaScript. Because it is generated from the same metadata the platform serves, it cannot drift from the real endpoint. This is one small fetch rather than a whole-catalog download, and it is the reliable way to get an exact schema. Endpoint IDs change as models are versioned, so prefer fetching this file over relying on an ID remembered from training data.

​
Reading a Documentation Page as Markdown

Any documentation page can be fetched as Markdown by appending .md to its URL:

​
Suggested Order

  1. Start at fal.ai/llms.txt to learn what the platform does and which models fit the task.
  2. Find a specific model in the model gallery or the documentation index.
  3. Fetch that model’s llms.txt for the exact schema, then call it using the client setup guide.