Show InvokeAI fleet status (version, queue, model counts) as a rich card.

Return Format

ToolResult with PrefabApp card; plain text fallback in content.

Examples

show_invokeai_dashboard_card()

Show the generation queue status as a rich card.

Return Format

ToolResult with PrefabApp card; plain text fallback in content.

Examples

show_invokeai_queue_card()

Show installed models of a type as a rich card.

Return Format

ToolResult with PrefabApp card; plain text fallback in content.

Examples

show_invokeai_models_card(model_type="main")

Show the most recent gallery images as a rich card.

Return Format

ToolResult with PrefabApp card; plain text fallback in content.

Examples

show_invokeai_gallery_card(limit=6)

Browse, search, download, and manage generated images.

[RATIONALE] The gallery is a single searchable feed (gallery endpoint) with per-image actions; one portmanteau keeps discovery and mutation together.

Return Format

{"success": bool, "operation": str, "data": {...}, "message": str, "images": [{"image_name", "url", "thumbnail_url", "width", "height"}]}

Examples

invokeai_gallery(operation="list", limit=20) invokeai_gallery(operation="search", query="cyberpunk") invokeai_gallery(operation="download", image_name="abc123.png")

Notes:

• URLs returned are absolute (InvokeAI host), usable in chat/browser.

Manage boards (collections of generated images).

[RATIONALE] Boards are the primary organization surface of the InvokeAI gallery; all board operations share one record store and one portmanteau.

Return Format

{"success": bool, "operation": str, "data": {...}, "message": str}

Examples

invokeai_boards(operation="list") invokeai_boards(operation="create", board_name="Concept art") invokeai_boards(operation="add_image", board_id="board-uuid", image_names=["abc123.png"])

Notes:

• Board ids come from invokeai_boards(operation='list').

Generate images through the local InvokeAI creative engine.

[RATIONALE] All generation paths share one enqueue flow (graph build -> queue batch), so they live under a single portmanteau with an operation discriminator instead of four near-identical tools.

• txt2img: text prompt to a fresh image.

• img2img: transform an existing image (needs image_name, uses strength).

• inpaint: regenerate a masked region (needs image_name + mask_image_name).

• upscale: RealESRGAN 4x upscale of an existing image (needs image_name).

The job is enqueued and runs asynchronously; poll completion with invokeai_queue(operation="item_status" or "result").

Return Format

{"success": bool, "queue_item_id": int, "batch_id": str, "message": str, "poll": {"tool": "invokeai_queue", "args": {...}}}

Examples

invokeai_generate(operation="txt2img", prompt="neon cyberpunk city at night, rain") invokeai_generate(operation="img2img", prompt="make it a watercolor painting", image_name="abc123.png", strength=0.6) invokeai_generate(operation="upscale", image_name="abc123.png")

Notes:

• InvokeAI must be running (onboarding) and a main model installed.

• Model base determines graph family: sd-1, sdxl, flux.

Manage installed models in the local InvokeAI instance.

[RATIONALE] Model lifecycle (discovery, install from HF/Civitai, config, removal) is a single domain with a shared record store, so all operations live under this portmanteau.

Sources accepted by install: HuggingFace repo ids (e.g. "stabilityai/stable-diffusion-xl-base-1.0"), Civitai model URLs, or local paths. Install runs asynchronously; poll with operation='installs'.

Return Format

{"success": bool, "operation": str, "data": {...}, "message": str}

Examples

invokeai_models(operation="list", model_type="main") invokeai_models(operation="install", source="stabilityai/stable-diffusion-xl-base-1.0", config={"name": "SDXL Base"}) invokeai_models(operation="delete", key="sdxl-main")

Notes:

• model_type values follow InvokeAI taxonomy (main, lora, vae, ...).

• Install may download multiple GB; check operation='installs' for progress.

Manage the InvokeAI generation queue and retrieve results.

[RATIONALE] Queue lifecycle (inspect, control, and harvest outputs) is one domain, so all operations share this portmanteau. 'result' is the natural partner of invokeai_generate: it polls an item to completion and returns the output image URLs (or downloads the file locally).

Return Format

{"success": bool, "operation": str, "data": {...}, "message": str}

Examples

invokeai_queue(operation="status") invokeai_queue(operation="list", status_filter="completed", limit=10) invokeai_queue(operation="result", item_id=123, wait_seconds=120, download_image=True) invokeai_queue(operation="cancel_batch", batch_ids=["batch-uuid"])

Notes:

• result polls every 3s up to wait_seconds; use wait_seconds for synchronous flows.

• Without wait_seconds, result returns the current item state immediately.

Check InvokeAI connectivity, version, and runtime configuration.

[RATIONALE] System introspection (health, version, config, cache stats) is one domain and the standard first call for onboarding and debugging.

Return Format

{"success": bool, "operation": str, "data": {...}, "message": str, "configured": bool}

Examples

invokeai_system(operation="health") invokeai_system(operation="version")

Notes:

• health returns configured=false when InvokeAI is not reachable (drives the webapp onboarding cue).

Get documentation for this server's tools and workflows.

Return Format

{"success": bool, "help": str}

Examples

invokeai_help() invokeai_help(topic="tools")

Gracefully shut down this MCP server.

Return Format

{"success": bool, "message": str}

Examples

invokeai_shutdown()

Manage InvokeAI node workflows (list, get, save, delete).

[RATIONALE] Workflows are stored artifacts in a single store; CRUD plus export share this portmanteau. Running a workflow graph goes through invokeai_generate or a raw enqueue - this tool manages the stored workflow library.

Return Format

{"success": bool, "operation": str, "data": {...}, "message": str}

Examples

invokeai_workflows(operation="list") invokeai_workflows(operation="get", workflow_id="wf-uuid") invokeai_workflows(operation="save", workflow_json='{"nodes": {...}}')

Notes:

• workflow_json must be a full InvokeAI workflow object (id, nodes, edges).

• Save with an existing id updates; without one, creates.