---
name: framemux
description: Discover FrameMux AI video, image, and audio models, estimate costs, and create or track authorized generations through its MCP server.
---

# FrameMux

Use the FrameMux MCP server to discover currently available models, inspect their input contracts, estimate costs, and create or track authorized generations.

## Connect

Add the remote Streamable HTTP MCP server. The client will open FrameMux OAuth authorization when an agent requests personal or generation tools:

`https://api.framemux.com/mcp`

For Codex:

`codex mcp add framemux --url https://api.framemux.com/mcp`

For Claude Code:

`claude mcp add --transport http --scope user framemux https://api.framemux.com/mcp`

## Workflow

1. Use `list_models` to find models that are currently available. Filter by `video`, `image`, or `audio` when useful.
2. Use `get_model` to read supported modes, parameter names, value constraints, and required reference inputs.
3. Use `estimate_generation` before text-only requests. Estimates are informational and do not charge credits.
4. Explain the current estimate and ask the user to confirm before calling `create_generation`.
5. Pass a stable idempotency key to `create_generation`; reuse it after timeouts or retries.
6. Use `get_generation` or `list_generations` to follow progress and return result links. Failed jobs are refunded automatically.
7. Reference-file uploads are not available through MCP yet; only use text input modes.

Never invent model parameters. Prices and availability may change, so request current information for each task.

More details: https://framemux.com/agents.md
