Tools, prompts, and resources
This page summarizes the server's catalog; your client always sees the current parameters through tools/list, prompts/list and resources/templates/list.
Tools
Finding and running services
| Tool | Purpose |
|---|---|
find_services | Ranks services by speed, cost and success rate. Takes a natural-language query ("fast cheap anime image", "best video") or an exact service id, an optional category (image, video, audio, chat) and limit. Also lists OpenRouter chat models, which run as the chatCompletion step. |
get_input_schema | The JSON Schema and an example input for a service. Pass a service id from find_services; or browse with stepType and parameters (open choices come back as options, with the services under them); or, for imageGen, pass Civitai checkpoint and LoRA AIRs as resources to get the matching input with them filled in. With no arguments it lists the step types. |
run_step | Runs one step: stepType and input as get_input_schema returns them. Returns each step's status and output, with resource links to any files it produced. |
run_workflow | Runs several steps in the format of POST /v2/consumer/workflows: {"$type", "input", "name"}, where $type is the stepType. A later step can take an earlier one's output with {"$ref": "<step name>", "path": "output.images[0].url"} in place of a value, and add "format": "A drawing of {0}" to insert it into text (details). |
get_guide | The plan for a multi-step job. Topics: civitai_models, image_to_video, edit_image, compare_services, understand_media. |
The recipes describe each step's inputs and outputs in more detail; get_input_schema returns the same shapes.
Understanding media
For agents that can't see or hear files themselves.
| Tool | Purpose |
|---|---|
caption_media | Describes an image or video. |
transcribe_audio | Transcribes audio or video to text, with optional word-level timestamps. |
Tagging (wdTagging) and content rating (mediaRating) run through run_step like any other step.
Workflows
| Tool | Purpose |
|---|---|
get_workflow | A workflow's status and each step's output, with resource links to files. |
list_workflows | Your recent workflows. Supports take, tags and excludeFailed. |
cancel_workflow | Cancels a running workflow. |
You can only see and cancel your own workflows.
Prompts
Prompts are plans your MCP client can offer you, usually as slash commands. Each returns the same plan as the matching get_guide topic, with your request filled in.
| Prompt | Arguments | Plan |
|---|---|---|
generate_with_civitai_models | prompt, resources (AIRs, comma-separated) | Get the input for exactly these models, add the prompt, run it. If the models can't run together, say why instead of swapping one. |
image_to_video | idea, optional image | Pick a video service that takes a source image; generate the first frame if none is given, and chain frame and video in one workflow. |
edit_image | image, instruction | Find a service that edits and describe the change plus what must stay the same. |
compare_services | prompt, optional category | Run one prompt on three different services in one workflow and compare cost, time and result. |
understand_media | mediaUrl, optional question | Caption or transcribe first, and add tagging or rating only when the question needs them. |
Resources
| URI template | Behavior |
|---|---|
spine://blobs/{blobId} | Images come back inline as base64. Videos and audio return a signed download URL valid for 5 minutes. Returns an error if the blob doesn't exist. |
Tools that produce media include resource links to this template, so MCP clients can show outputs inline without a separate download.
Related
- MCP Server overview: endpoint, authentication and client setup
- Recipes: REST equivalents with runnable examples
- API Reference: generated from the OpenAPI spec