← Back to blog
use case·10 min read·July 30, 2026
Leia em português →Leer en español →Auf Deutsch lesen →Lire en français →

How to Add Video Generation to Your AI Agent

Give an AI agent the ability to generate images and video through the 8frame MCP server. Why MCP beats raw model APIs, architecture patterns, and a human-in-the-loop design.

Correction, 2026-09-30: the first version of this page said the MCP server picks the model for you (a model is chosen per generation, from a picker with credit costs), priced models in dollars that don't match 8frame's credits, and gave a fixed generation time we can't source. This version corrects those.

Correction, 2026-10-01: prices are now the connector's own. Through the connector, Kling runs as Kling v3 Standard with sound and needs a start image (86 credits for 5 s; the 57-credit silent clip is canvas-only), Veo 3.1 has no silent setting (448 for 8 s on Standard, its default), and the Nano Banana tool is the 6-credit model, so the four-clip example is 344 credits, not 228. Also corrected the same day: the page said a direct model API is cheaper only at massive, negotiated scale (Google's list price for Veo 3.1 Standard, $3.20 for 8 seconds, is already below 8frame's 448 credits), that your agent's tool interface stays stable when models change (connector tools are sometimes retired), and that agent output lands on the canvas (the connector returns file links). It called the three-stage loop the pattern that "holds up in production", which we haven't measured, counted the connector's tools as models, asked you to sign in to a "workspace" (you sign in to your 8frame account), and showed a 9:16 Kling example without saying that the start image sets the shape.

You've built an agent that drafts campaign briefs, or triages support tickets, or writes launch copy. The next thing someone asks for is visuals: "can it also make the video?" The clean way to add that is not to wire your agent to a dozen individual model APIs. It's to give it one tool server that speaks MCP and let that server handle image generation and editing, background removal, image upscaling, video and music. This guide covers why that's a sensible integration layer, an architecture pattern to start from, and how to keep a human in the loop where it matters.

By the end you'll understand how an AI agent can go from "draft the brief" to "generate the assets" without you maintaining a pile of per-model client code, and what a sane approval flow looks like around it.

What you're actually adding

An agent generates video by calling a tool. The question is what sits behind that tool. Two options:

  1. Raw model APIs. You integrate each provider directly: one client for a Veo endpoint, another for Kling, another for an image model, another for an upscaler. Each has its own auth, request shape, error handling, and rate limits.
  2. One MCP server. You point your agent at a single server that already exposes image generation, editing, upscaling and video across many models behind one interface.

The 8frame MCP server is the second option. Server URL https://mcp.8frame.co/mcp, exposing 8 image-generation tools, 9 video tools and 2 music tools (one per model family, several of them with more than one model or tier inside), plus image editing, background removal and image upscaling (as of 2026-10-01). Your agent gets creative generation as a capability; you don't maintain the plumbing.

Why MCP beats raw model APIs here

One integration, not N. MCP is an open standard for connecting AI applications to tools and data, introduced by Anthropic in November 2024. The protocol's own site lists support in AI assistants such as Claude and ChatGPT and in developer tools such as VS Code and Cursor. If your agent framework or client speaks MCP, you add one server instead of writing and maintaining a client per model.

Model changes mostly stay behind the server. When a new video model lands or a provider changes its API, that work happens behind the MCP server, and your agent's calls don't need rewriting. Tools do change now and then: when a provider shut down a video model in September 2026, the connector removed that tool. Have your agent read the tool list when it connects instead of hard-coding tool names. With direct integrations, every provider update is your problem.

Model choice stays explicit. Each model family is its own tool, and two "pick a model" tools show the options with their credit costs, in clients that support that kind of prompt. One model is picked per generation, by the user or by your agent's rules (cheap model for variants, Veo for the hero). Your agent doesn't need a client per provider to make that choice.

One cost model. Everything is billed in 8frame credits from one balance. At the connector's prices: Kling v3 Standard 86 credits for 5 seconds with sound, Veo 3.1 448 for 8 seconds with sound on Standard (its default), 168 on Fast and 54 on Lite, Pruna p-video Draft 4 for 5 seconds, Nano Banana 6 an image. A credit is $0.01 at pack rate. The connector has fewer settings than the canvas: no silent option on Kling or Veo, and no length setting on either. Your agent budgets in one unit instead of a price map per provider.

The honest tradeoff: raw APIs give you the lowest-level control, and at list price a direct API can be cheaper per clip. Google's Gemini API lists Veo 3.1 Standard at $0.40 a second with audio, $3.20 for 8 seconds, against 448 credits (about $4.48) through 8frame. If your product is only video generation at high volume on one model, integrate that model directly. For an agent that needs creative generation as one capability among many, MCP is usually the better fit.

How to connect it

The server is a standard remote MCP endpoint, so any MCP-capable client or framework attaches to it the same way. In Claude Code, for example:

claude mcp add --transport http 8frame https://mcp.8frame.co/mcp

Then run /mcp inside Claude Code and sign in to your 8frame account in the browser (the server uses OAuth). In an agent framework that supports MCP, register https://mcp.8frame.co/mcp as a remote tool server in your config and complete the same sign-in; how a framework handles OAuth varies, so check its docs. The generate, edit, upscale and video tools then become available to the agent's planner, and generations are charged to the credits of the account you signed in with. Setup notes are at 8frame.co/mcp. If your agent runs inside claude.ai rather than your own stack, the Connectors setup is the equivalent route.

Architecture pattern: agent drafts, MCP generates, human approves

A pattern worth starting from is a three-stage loop, not a fully autonomous one:

Stage 1: Agent drafts the brief. The agent does what it's good at: turning a goal ("launch creative for the new SKU, three hook variations") into structured prompts. Subject, action, camera, lighting, aspect ratio, one prompt per variation. This is text, so it's cheap and reviewable.

Stage 2: MCP generates. The agent calls the matching 8frame tool with each prompt and the chosen model, and results come back as file links (not canvas boards). For a variation set, the agent fans out: three or four prompts, three or four clips, 258 to 344 credits on Kling v3 Standard (86 a clip).

Stage 3: Human approves. The generated set surfaces to a person, who approves, flags, or requests a revision. Only approved assets move to publish. This is the guardrail that keeps a stochastic generation step from shipping something wrong.

That loop mirrors how the agent pattern works generally: goal in, tool calls, evaluation, and a stopping condition that hands off to a human when judgment is required. Video generation is exactly the kind of step you don't fully automate, because the failure mode (a warped product, an off-brand look) is visual and subjective.

Two design decisions that matter

Put the human before publish, not before every generation. A clip nobody uses costs only its credits, and the fan-out cap below bounds that. So let the agent generate, collect the returned files for review, and put the approval gate between "generated" and "delivered to the client or ad account," where a mistake costs more than credits. If you want an earlier check, approve the drafted prompts: they are text and quick to review.

Cap the fan-out. An agent that can call a generation tool in a loop can also spend credits in a loop. Set a per-run generation limit: a brief that should produce four variations shouldn't be able to trigger forty because of a planning error. Pin the model tier as well, since a Veo call that names no tier runs Standard at 448 credits. 8frame also caps the credits spent through the connector per user per day as a safety limit; treat it as a backstop, not as your budget.

A worked example

An agent handling paid-social creative for a DTC brand:

  1. Input: "New hook test for the sleep supplement, four angles, vertical."
  2. The agent drafts four prompts, each a different UGC hook angle (problem-first, product-first, testimonial, before-after), all vertical.
  3. It calls the 8frame Kling tool (Kling v3 Standard) four times, with a vertical 9:16 product photo as the start image. With a start image, the Kling call sends the model no aspect ratio, so the photo sets the shape. Four 5-second clips with sound cost 344 credits, about $3.44.
  4. The set surfaces to the performance marketer, who approves two, kills one, and asks for a re-roll on the fourth with a tweaked prompt the agent supplies. The re-roll is one more 86-credit clip, so the run comes to 430 credits, about $4.30.
  5. Approved clips route to the ad account.

The agent did the tedious part (drafting and generating variations); the human did the judgment part (which hooks are worth spending on). That division is the whole point.

Limits to design around

Latency is real. A video generation takes from about a minute (a 5-second Kling v3 Standard clip, median on 8frame) to about five minutes (Seedance 2.5 at 720p, the version the connector runs, median on 8frame). The connector waits up to 10 minutes, then returns a task ID that its check tool polls. Treat these as async tool calls and report progress rather than hanging.

Review can't be skipped. The three-stage loop exists because generation is probabilistic. An agent that auto-publishes will sooner or later publish a bad frame. Keep the human gate.

Fine control still lives on the canvas. The connector exposes fewer settings than the canvas: clip length and silent mode on Kling and Veo, end frames, Kling v3 Pro and Nano Banana Pro are canvas-only. For shots that need those, or precise composition from many reference images, an operator opens the canvas directly. The agent handles volume and drafting; the canvas handles the shots that need hand direction.

FAQ

How do I connect an agent to 8frame?

Register https://mcp.8frame.co/mcp as a remote MCP server in your client or framework and sign in to your 8frame account. In Claude Code: claude mcp add --transport http 8frame https://mcp.8frame.co/mcp, then /mcp to sign in.

Does the MCP server choose the model?

No. Each tool runs one model family, and the "pick a model" tools show the options with credit costs; your agent's own rules can choose too. When a call names no tier, the tool's default applies: on Veo that is Veo 3.1 Standard, 448 credits.

What does a four-clip variation set cost?

Through the connector, four 5-second Kling v3 Standard clips are 344 credits, about $3.44 at pack rate; the connector's Kling runs with sound and needs a start image. On the canvas, the same four clips silent are 228.

Sources

Related reading

To understand the protocol your agent is speaking, start with what is MCP. For the prompt-by-hand version of this setup, see how to generate AI video from Claude or, for developers, AI video in Cursor and Claude Code.

Give your agent creative generation through one server. The 8frame MCP setup is live, or explore the canvas and workflow library for the shots you direct by hand.

Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.

Written with AI assistance. Corrections are noted on the page. Editorial policy

Related articles

use caseAI Video in Cursor and Claude Code via MCPuse caseHow to Generate AI Video from Claude (MCP Guide)use caseHow to Make an ASMR Product Video with AI

Make it
move.

Stay in the loop

Be the first to hear about our launch and get product updates