# FrameMux for agents

FrameMux provides AI video, image, and audio generation with one credit balance.

## MCP server

- URL: https://api.framemux.com/mcp
- Transport: Streamable HTTP
- Authentication: model discovery and estimates are public. Generation and account tools require OAuth authorization in the MCP client.
- MCP endpoint: send MCP JSON-RPC requests to the URL above.

## Available tools

- `list_models`: find available models, optionally filtered by kind (video, image, audio) or search query.
- `get_model`: inspect a model's supported modes, parameter constraints, and reference requirements.
- `estimate_generation`: get a live estimate for text-only generation. This does not create a job or charge credits.
- `create_generation`: create a text-only job using the authorized user's credits. This tool requires the `generation:create` scope.
- `get_generation` and `list_generations`: read jobs owned by the authorized user. These tools require the `generation:read` scope.

## Workflow

1. Call `list_models` to discover currently available models.
2. Call `get_model` before selecting parameters. Do not invent parameter names or values.
3. Call `estimate_generation` with the model key, prompt, and supported parameters. Estimates can change before generation.
4. Before `create_generation`, explain the live estimate and get the user's confirmation. Include a stable idempotency key and reuse it for retries.
5. Poll `get_generation` for completion and return result links. Failed jobs are refunded by FrameMux.
6. Reference-file uploads are not available through MCP yet.

## Links

- For Agents: https://framemux.com/for-agents
- Model catalog: https://framemux.com/models
- Pricing and credit rules: https://framemux.com/pricing
- Official Skill: https://framemux.com/skills/framemux/SKILL.md
