MCP image and video generation in OfflineCreator Studio
Use OfflineCreator MCP to list models, check credits, submit an image or video job, wait, and download the result. This is hosted cloud generation, not on-device.
Use OfflineCreator Studio MCP when a creator wants an MCP client to inspect model costs, submit an image or video job, follow its status, and retrieve the result without wiring separate provider APIs. The recommended remote connection is the hosted Streamable HTTP endpoint with browser OAuth. Its live resource advertises models, read, and generate scopes, and the published client documentation says remote OAuth keeps an API key out of the client configuration.
This is a cloud workflow, not local generation. Studio sends generation work to disclosed cloud providers; creators who require on-device processing should take the LocalForge route instead. That boundary matters more than whether the request starts in Cursor, Claude, or another compatible client.
Keep the cloud and local boundaries explicit
MCP changes how an agent reaches generation tools; it does not make the generation local. Studio MCP and CLI generations use disclosed cloud providers, while LocalForge is the fully offline path. Do not describe Studio as offline merely because the interaction happens inside a desktop client.
What the nine MCP tools cover
The current package exposes nine tools rather than one oversized generation call. list_models and get_credits support a cost check; generate and upload_input start text-led or reference-led work; get_generation, wait_generation, and list_generations cover job tracking; download_output returns a completed result; cancel_generation stops a reserved job. Tool visibility is filtered by the authorized models, read, or generate scope.
That surface supports a bounded creator workflow: inspect the catalog, check the balance, choose a model, submit, wait or poll, then retrieve the output. Image-to-video needs the upload step. Top-up discovery and checkout are CLI or API features, not MCP tools, and the published package keeps payment completion as a human browser action.
- Discover and budget
- list_models · get_creditsReview the workflow and published credit cost before a generation.
- Create
- generate · upload_inputStart text-to-image or text-to-video directly; attach a source image for image-to-video.
- Track and retrieve
- get_generation · wait_generation · list_generations · download_outputFollow the job and retrieve a completed output instead of assuming immediate completion.
- Stop
- cancel_generationCancel a reserved generation when the job no longer serves the brief.
Six launch models and their published credit costs
The current catalog has three image choices and three video choices. FLUX Schnell costs 1 credit for fast image iteration, FLUX 1.1 Pro Ultra costs 8 for higher-detail images, and Recraft V3 costs 5 for design-forward graphics. Kling 2.6 Pro costs 42 credits for text-to-video, Kling 2.6 Pro Motion costs 42 for image animation, and Veo 3.1 Fast costs 96 for premium text-to-video.
Treat these as a current catalog snapshot, not permanent model availability or a quality ranking. The useful decision is workflow first: choose text-to-image, text-to-video, or image-to-video; compare the visible cost with the remaining balance; then reserve credits only after the prompt, aspect ratio, and source rights are ready.
- Text-to-image
- FLUX Schnell 1 · FLUX 1.1 Pro Ultra 8 · Recraft V3 5Use low-cost iteration before moving to a more expensive final candidate.
- Text-to-video
- Kling 2.6 Pro 42 · Veo 3.1 Fast 96The published costs differ sharply, so model selection changes the available iteration count.
- Image-to-video
- Kling 2.6 Pro Motion 42Prepare a rights-cleared still and use upload_input before submission.
Continue by the next decision
Compare hosted credits with a bring-your-own-key integration when credential ownership is the deciding factor. Read the job-lifecycle page when cancellation, status, and retrieval behavior matter. Use the creative-agent page for the broader orchestration decision, or return to the MCP and CLI hub for connection details.
Evidence and editorial boundary
This page owns the commercial decision for MCP image and video generation: tool coverage, launch models, current credit costs, pricing, and the cloud-versus-local fork. Keep connection syntax on client pages and deeper pricing mechanics on the pricing page. Recent community coverage was insufficient and does not substantiate OfflineCreator reliability, speed, quality, adoption, or customer outcomes.
The complete evidence refresh retained 53 item-level records: 15 Digg clusters, 13 GitHub records, three general-web results, three Hacker News stories, two Instagram reels, six Reddit threads, ten TikTok videos, and one YouTube video. Item review classified 12 as marginal context, 10 as promotional, 28 as irrelevant, two as unverifiable hashtag-only videos, and one as unsafe because it asserted conversion performance without evidence. None directly verified OfflineCreator's nine-tool surface, six-model catalog, pricing, reliability, quality, speed, or customer outcomes, so no community source or claim was added.
Coverage remains degraded rather than silent. Reddit returned six records before HTTP 429 and X was unavailable because optional credentials were not configured. The successful media lanes mainly surfaced competitor tutorials, promotions, generic MCP introductions, and unrelated model announcements. The single YouTube walkthrough had 19 views and no engagement, while several promotional posts made workflow or performance claims that were not independently tested. Existing first-party product, package, pricing, and OAuth metadata remain the stronger evidence for this page.