MaxVideoAI

MaxVideoAI for assistants, agents, and automations

MaxVideoAI is a multi-model AI production service exposed through one remote MCP server. It gives assistants, agents, and approved automations a controlled path to plan video and image work, compare current models, budget projects, prepare exact quotes, approve one paid attempt, recover results, and keep work in the connected MaxVideoAI Library.

Plan. Compare. Price. Approve. Generate. Recover.

Setup guides: Claude · ChatGPT · Codex · OpenClaw · n8n — see current compatibility and protocol details.

[Public MaxVideoAI MCP integration grid for Claude, ChatGPT, Codex, OpenClaw, and n8n on a light background]

Plan a production · Compare models · Review pricing

Install the repository-validated Codex package

codex plugin marketplace add camgraphe/maxvideoai-plugin --ref v0.3.8codex plugin add maxvideoai@maxvideoai

Start a new Codex task, then ask $maxvideoai:plan to compare a production route or $maxvideoai:generate to prepare a concrete request. The plugin is free to connect. The first live MaxVideoAI action opens OAuth so you can sign in or create the account you want to connect.

Public releases include a versioned ZIP and SHA-256 file. Match the selected public tag to VERSION before installation.

What does MaxVideoAI add to an AI assistant?

The assistant owns the creative conversation; MaxVideoAI supplies current product truth and the controlled execution path. It can read live model details, compare compatible options, calculate complete project budgets, select account-owned media, validate concrete settings, return an exact quote, submit one approved attempt, follow the accepted job, and present the finished result.

[Public MaxVideoAI home page with a colorful MiniMax H3 Max film preview and pay-as-you-go creation actions]

The same connected account and Library are used on the website and through the plugin. There is no separate plugin subscription; approved generations use existing MaxVideoAI credits at current pay-as-you-go prices.

Which setup should you use?

• Claude: add the remote connector/plugin, complete OAuth, then begin with a no-spend plan. Follow the Claude guide.
• ChatGPT: connect the direct MCP in developer mode and complete OAuth on first use. MaxVideoAI is not submitted to the OpenAI directory under the current commerce policy. Follow the ChatGPT guide.
• Codex: install the tagged public package, open a new task, and call $maxvideoai:plan or $maxvideoai:generate. Follow the Codex guide.

Agent and automation paths

Direct host paths: Claude · ChatGPT · CodexAgent and automation paths: OpenClaw · n8n

• OpenClaw: the remote MCP path is tested with limits around private references, channel attachments, and inline channel rendering; the packaged skill is listed on ClawHub. Follow the OpenClaw guide.
• n8n: on tested self-hosted n8n, use the deterministic MCP Client workflow with manual configuration. n8n Cloud and MCP Client Tool agent invocation are not claimed. Follow the n8n guide.
• Another MCP client: connect only if the client supports remote Streamable HTTP and OAuth. Follow the generic MCP guide.

Every route connects to https://api.maxvideoai.com/mcp. Add that URL exactly as written—without a token, query string, password, or API key.

Cursor, GitHub Copilot, Gemini CLI, and Microsoft Copilot are in preparation. They do not yet have public MaxVideoAI setup paths.

Connect → OAuth on first use → plan without spending → prepare quote → approve once → recover result

What can you create?

Start from a brief, a supported private reference, or a public example. The plugin can help turn that material into a shot plan and a model-aware request, while the MaxVideoAI gallery gives you concrete prompts, settings, durations, and current price estimates to inspect before building your own version.

[MaxVideoAI examples gallery with brand filters and four opening video cards for Wan, Seedance and FLUX]

Explore AI video examples or copy one of the four practical agent workflows.

How do you compare models side by side?

Use $maxvideoai:plan in Codex or /maxvideoai:plan in Claude Code. Planning reads current model facts and builds comparable, named production routes without authorizing generation. Ask for one consistent-model route, a deliberate model mix, a quality-focused plan, or a credible lower-cost alternative using the same shot assumptions.

[MaxVideoAI comparison page showing MiniMax H3 and Seedance 2.5 scores, strengths, and colorful video examples]

The engine comparison hub and model directory expose the current decision surface. A recommendation is a capability match for the brief, not a guarantee of provider availability or a paid quote.

How do pricing and approval stay clear?

A project budget is useful while you are choosing a direction. When the model, prompt, settings, and required references are concrete, $maxvideoai:generate or /maxvideoai:generate prepares the exact price for that request. Nothing is submitted as paid work until you explicitly approve that quote, and the approval covers one attempt.

[MaxVideoAI pricing page with three current video starting points and navigation across video, image, audio, and tools]

1. Plan: compare current model options and complete project budgets without spending credits.
2. Prepare: validate the selected request and supported private references.
3. Price: receive the exact quote for those settings.
4. Approve: explicitly authorize one paid attempt.
5. Recover: follow the accepted job before considering another submission.

Which preparation and finishing tools are available?

The public MaxVideoAI toolbox covers repeatable work around generation: character sheets, camera-angle exploration, image refinement, background removal, and video finishing. It keeps the next production step visible without implying that every tool is part of this plugin package or available through every MCP host.

[MaxVideoAI tools page showing a colorful character sheet, camera-angle work, image refinement, and video finishing options]

Use the live tool page for current inputs and prices. The remote MCP exposes only the capabilities in its current tool contract.

The current live MCP surface covers video and image planning, quoting, generation, recovery, and private-reference handling. The owner-approved 17 October 2026 server release adds bounded Eleven v4 narration through the independent Audio gate. Studio montage remains gated. Installed package 0.3.7 routing and dated host evidence are separate from that server release; refresh the connected client's tool inventory before using new tools. Trust that inventory over screenshots, examples, or remembered capability names.

How do private references and the Library work?

Compatible ChatGPT or Claude surfaces can import authorized attachments and generated results as private MaxVideoAI assets without a public URL or Computer Use. Codex and Claude Code use the packaged local helper when a host cannot expose a temporary file handle directly. The helper reads local bytes itself; it does not publish the file or pass a raw local path to the MCP server.

[Public MaxVideoAI MCP page showing a finished colorful video in Claude with confirmation that it was saved to the connected Library]

Private references and finished generations remain in the connected MaxVideoAI Library. If a response is interrupted after submission, recover the accepted job or recent generation before preparing another paid attempt. Read privacy and permissions and reference input handling for the complete boundary.

Open the MaxVideoAI Library to continue with saved results and private media.

Try asking

Compare current AI video models for a 20-second product film. Give me aquality-focused route and a credible lower-cost route using the same shotassumptions. Do not prepare paid work yet.
Use my authorized product image as the first frame where the selected workflowsupports it. Prepare the exact quote, but wait for my explicit approval.
The conversation stopped after approval. Check the accepted job before youconsider another paid submission, then recover the result in my MaxVideoAI Library.

More workflows:

• Compare current AI video models
• Price an AI video project
• Plan a Claude production
• Run a Codex production workflow

How is the plugin packaged?

The repository is deliberately reviewable. It contains host manifests, one remote MCP endpoint, two scoped skills, human setup guides, producer examples, community files, release graphics, and a deterministic public-bundle builder.

.mcp.json                         remote MCP connection.claude-plugin/                   Claude package metadata.codex-plugin/                    Codex package metadataskills/plan/                      model comparison and project budgetingskills/generate/                  quote, approval, execution and recoverydocs/                             host setup, privacy and troubleshootingexamples/                         copyable production workflowsscripts/import-reference-files.mjs local private-file helper

The MCP server uses OAuth and keeps pricing, account data, private media, jobs, and provider execution on the MaxVideoAI side. The host keeps the creative conversation and prompt development. Read how it works for the complete division of responsibility.

How do you validate a contribution or release?

From the source repository root:

pnpm github:content:checkpnpm github:assets:release-checknode --test --import tsx tests/github-content-contract.test.tsnode --test --import tsx tests/mcp-public-release-bundle.test.ts

The release builder exports an exact allowlisted file set, validates every referenced image against the asset manifest, rejects secrets and unsafe paths, writes SHA-256 checksums, and produces the versioned archive used by the dedicated public repository.

See CONTRIBUTING.md for review scope, SECURITY.md for private vulnerability reporting, and SUPPORT.md for product and compatibility help.

Policies and product help: privacy · terms · contact · [email protected] · Business Source License 1.1.

Last reviewed: 2026-09-16.