Scenario MCP skill set: have AI agents generate images, video, audio, and 3D

Who this is for

• Anyone who wants an AI coding agent to directly produce production-ready images, video, audio, and 3D content: the official positioning is to draw on this capability from any AI agent, with the skills picking the right model, quoting a price before spending, and keeping characters and brands consistent; no matter which product the agent comes from, generation is delivered through the same remote MCP endpoint.
• Art and development teams whose workflow revolves around Blender, Maya, ZBrush, Unreal, or Unity: the repository description also mentions expert capabilities that drive these tools, which fits wiring an agent into an existing production pipeline; it also fits users who already have concrete asset goals and want to handle model selection and creation inside the agent.
• Prerequisite: content generation goes through the remote Scenario MCP. Connecting to https://mcp.scenario.com/mcp requires signing in with a Scenario account via OAuth, or using an API key instead — one or the other; without credentials there is no way to reach the generation capability.
• What it is: not a standalone server process, but a SKILL.md skill file that loads on demand, as procedural knowledge, when the agent needs it; if your client cannot add a remote MCP endpoint, the generation path leads nowhere. Pricing and how to obtain an API key are not explained on the official page, and for users who simply want to click and generate in a browser without an agent, the official page offers no such path either.

First attempt in 3 minutes

• Prerequisites: a local AI coding agent that can install skills, plus Scenario credentials ready — OAuth sign-in or an API key, one or the other; OAuth sign-in requires an existing Scenario account.
• Step one, install the skills (a local write): run npx skills add scenario-labs/skills --skill scenario in a terminal. This step writes skill files into the local agent environment; it is not a read-only operation. Per the official documentation, installs always fetch the latest copy of each skill from main, with no version pinning.
• Step two, connect the MCP (remote configuration): run claude mcp add --transport http scenario https://mcp.scenario.com/mcp to register the server as scenario; or add https://mcp.scenario.com/mcp to any MCP client, then sign in with a Scenario account (OAuth) or use an API key instead. Both ways of connecting come from the official documentation; this guide treats only the CLI command as the primary path.
• Step three, a low-risk first try: do not have the agent generate assets right away. Let it pick a suitable model based on the installed skills and quote a price before spending — the official documentation stresses “price before spending”, which tells you that real generation costs money. At this step, observe the model and price the agent proposes first, and only after seeing them decide whether to proceed to generation.
• Success signal: the install command completes without errors, and a server named scenario appears in the MCP client with sign-in or authentication passing. If the install errors out or the scenario server cannot be found in the client, rerun the install command first, then check whether sign-in or the API key was configured successfully. Skill installation is a local state write, and the official documentation provides no uninstall or rollback path.

Use cases

• Produce new material while keeping characters and brands consistent: the project already has a set of character or brand visuals and needs more images or video. After installing the scenario skill and connecting the MCP, describe what you need in the agent, let it pick a model and quote a price first, and after your confirmation generate content that stays consistent with your characters and brand; you can expect production-ready images, video, audio, or 3D output, and you will see the price before any generation happens. If the price does not suit you, simply stop at the quote step — you have not yet entered the generation action that incurs cost.
• Connect Scenario into an existing MCP client: your main environment is another client that supports remote MCP, and you do not want to switch tools for this capability. Beyond CLI registration, the official documentation says you can add https://mcp.scenario.com/mcp to any MCP client, then sign in with a Scenario account (OAuth) or use an API key instead. After connecting, you can confirm that the endpoint has been added and that sign-in or authentication passes; the agent in that client then draws the generation capability through the Scenario MCP. Model selection and quoting a price before spending are handled by the skills; for skill installation, see the command in the previous section. The menu steps for adding the endpoint in each client, and the specific tools behind the endpoint, are not documented by the official page.

Boundaries

• Read/write nature: skill installation is a local state write; content generation is a service call over the remote MCP, and the official documentation stresses “price before spending” — it incurs cost, and the pre-generation price confirmation is exactly the confirmation point before the action.
• Credentials: connecting to https://mcp.scenario.com/mcp requires Scenario account OAuth sign-in or an API key — one or the other; how to apply for and configure an API key is not explained on the official page.
• Versioning: installs always fetch the latest copy from main, with no version pinning, so the same command may install skills with different content at different times; how to lock a version is not explained on the official page.
• Expert skills: the repository description mentions expert capabilities that drive Blender, Maya, ZBrush, Unreal, and Unity, but the only install command documented on the official page is npx skills add scenario-labs/skills --skill scenario, whose --skill scenario points at the scenario skill; how these expert capabilities are installed and invoked is not documented on the official page and cannot be inferred from that command.
• Unverified items: the specific tools behind the MCP endpoint, pricing and billing, whether skills work offline, how each skill is loaded and triggered, and uninstall or rollback paths — none of this is explained on the official page. Not finding an explanation is not the same as there being no risk; this guide draws no safety conclusions of its own, and the official page and the client’s actual prompts are what count.