MaxVideoAI MCP

npx add-mcp 'https://api.maxvideoai.com/mcp'

MaxVideoAI connects compatible assistants and automation tools to the models, credits and private media library of your MaxVideoAI account. Use it to compare current video models, prepare a quote, approve generation and recover the result.

Choose your integration

Environment	Setup	What to know
Claude connector	Custom remote MCP connector	Claude Desktop tested; Claude Code has a separate, unverified host path.
ChatGPT MCP app	Developer mode	Full MCP write actions require an eligible Business, Enterprise or Edu workspace. Pro supports read/fetch only. No public directory listing is claimed.
Codex plugin	MaxVideoAI plugin 0.3.3 or remote MCP	Codex CLI tested.
OpenClaw skill	Direct MCP or listed ClawHub skill	Tested with limits; private-reference import and inline channel rendering are unverified.
n8n video workflow	Manual, self-hosted MCP Client	Tested on 2.38.7 with explicit approval. No public template listing or n8n Cloud support is claimed.

Cursor, GitHub Copilot, Gemini CLI and Microsoft Copilot are in preparation. Their MaxVideoAI setup pages are not published.

MCP server address

https://api.maxvideoai.com/mcp

Use Streamable HTTP with OAuth 2.1. The requested scopes are openid,email,profile. Do not append a token, password, query parameter or API key. Follow the guide for your client: it opens the MaxVideoAI sign-in and OAuth consent screen in your browser.

Authorize your account

1. Configure the connection using the guide above.
2. Sign in to MaxVideoAI or create an account, and confirm your email if requested.
3. Review the authorization, then return to the client and check the connected account.
4. Review and approve the exact quote before any paid generation.

Connecting is free and planning adds no MaxVideoAI subscription. Paid generations use the same MaxVideoAI credit balance as the website; your assistant provider may have its own plan requirements. Remove the integration from the client and revoke its authorization in MaxVideoAI to disconnect.

Compatibility is recorded per host. Claude Desktop and Codex CLI have verified checkpoints; ChatGPT web and Claude Code have no recorded end-to-end checkpoint. OpenClaw and the self-hosted n8n MCP Client retain their tested-with-limits status. See the integration guides for dates and restrictions.

What the assistant should do

The assistant remains responsible for the creative conversation: brief, script, shot plan, prompt writing, and reference ideas. It should use MaxVideoAI when it needs changing facts or an account action:

• call list_models and get_model_details instead of relying on model memory;
• ask only for choices that materially affect quality, budget, duration, audio, or references;
• respect a compatible model chosen by the user;
• when the user is undecided, present the best executable fit first and credible lower-cost alternatives;
• use calculate_project_budget to price comparable complete-film proposals with the same retry assumptions;
• use prepare_generation only when the final request is ready;
• wait for explicit approval of the returned exact quote before confirm_generation.

Quality is not a single preset. For multi-shot work, the assistant may compare a quality-first plan with one or more mixed-model alternatives, explaining the model selected for each shot. Creative attempts are billable scenarios and should be shown separately in the budget.

Tools

Tool	Purpose	Writes or spends?
get_account_status	Read connection, credit balance, trial state, limits, and safe account destinations.	No.
list_models	Find current executable image and video models by mode and requirements.	No.
get_model_details	Read exact settings, reference roles, limits, evidence, and current pricing inputs for one model.	No.
recommend_models	Match the brief to a short list of factual model options and trade-offs.	No.
calculate_project_budget	Compare one to four complete project proposals at current prices.	No. An estimate is not a quote.
list_media	List the connected user's ready private image, video, or audio assets.	No.
create_reference_upload_link	Create a short-lived private MaxVideoAI browser upload handoff.	Creates a handoff, never a provider job.
import_reference_files	Import up to eight user-authorized host attachments or generated file results into the private library.	Stores private assets; never starts a generation.
prepare_generation	Validate model, settings, references, ownership, and balance; return an exact short-lived quote.	No charge and no generation.
confirm_generation	Approve one exact quote and submit its job idempotently.	Yes, only after explicit approval.
get_generation_status	Read a known job, completion, output, failure, and refund state.	No.
list_recent_generations	Recover recent jobs instead of accidentally submitting duplicates.	No.
present_generation	Present a completed owned image or playable video inline in compatible hosts.	No. Uses the result link and library as fallback.
create_topup_link	Open a secure MaxVideoAI credit top-up handoff for the missing amount.	Payment happens only on MaxVideoAI.

Production publication flags decide which write-capable tools are exposed. A client must not infer availability from this reference alone; its live tools/list response is authoritative.

Images, video, and audio references

The assistant may help invent, create, or select reference media. MaxVideoAI accepts and stores the media but does not require the assistant to surrender its creative role.

Use list_media for an asset already in the connected MaxVideoAI library. When ChatGPT or another compatible host exposes user-authorized attachments or a generated file result, import_reference_files imports up to eight files and returns private assetId values ready for the request; no new library listing is needed.

If the host does not expose file handles, call create_reference_upload_link with image, video, or audio. A compatible ChatGPT or Claude surface can display its in-chat multi-file importer. The returned HTTPS browser page remains the manual fallback. Codex and Claude Code can use one link per local file with the packaged helper. No path sends a raw local filename to the MCP server, publishes a public URL, or requires Computer Use.

Always call get_model_details for the selected mode. It returns the accepted media kinds, roles, order, counts, formats, sizes, and duration limits. Do not copy reference rules from a different model. Private source URLs and stable assetId values stay inside the account boundary.

Seedance 2.5 is available through its working ModelArk route for text-to-video, image-to-video, reference-to-video, and extension. Direct video-to-video that depends on BytePlus LAS is not exposed until LAS access exists; that limitation does not disable the supported Seedance 2.5 modes.

Budget, exact price, and approval

calculate_project_budget is free and useful early. It estimates a film from named proposals, shot durations, model choices, and explicit creative-attempt assumptions. It does not reserve capacity, debit credits, or create a job.

When the user chooses a concrete request, prepare_generation performs final validation and returns the exact account price with an expiry. Display that price and wait for an unambiguous approval. Only then call confirm_generation once with the quote identifier.

If credits are insufficient:

1. call create_topup_link for that quote;
2. send the user to the returned MaxVideoAI billing page;
3. after the user says payment is complete, call get_account_status again;
4. call prepare_generation again because the previous quote is no longer valid;
5. show the fresh exact price and request approval again.

The plugin itself, model advice, and project budgets require no separate subscription. Paid generations use the existing pay-as-you-go MaxVideoAI credit balance.

Jobs, library, failures, and retries

An accepted job is not yet a completed video. Poll get_generation_status for a known job or call list_recent_generations to recover it. Once it is completed, present_generation can render the image or playable video inline in an MCP Apps-compatible host. Other clients keep the returned result link and MaxVideoAI library destination as the universal fallback. Do not submit the same paid generation again merely because the first call timed out.

Successful images and videos are saved to the same private MaxVideoAI account library used by the website. Return only the official result, workspace, or library destination supplied by MaxVideoAI.

For a definitive technical failure, explain the returned failure and refund state. Do not retry automatically. A creative retry is a new paid attempt and needs a new quote and a new explicit approval.

Introductory credit

An eligible verified account may receive one introductory Seedance 2 Mini generation when the promotion is active. This is a one-time promotional entitlement, not reusable wallet money. Its availability is returned by get_account_status; never promise it before checking the connected account. Audio may remain enabled without changing the promotional price when the live preset says so.

Disconnect and revoke

1. Remove MaxVideoAI from the client.
2. Open MaxVideoAI account connections.
3. Revoke the relevant grant.
4. Confirm that the next protected call requires a fresh browser approval.

Privacy and security

• The account email is omitted from tool results.
• Prompts, tokens, secrets, full reference URLs, private media, and payment details are excluded from acquisition analytics.
• Authenticated responses use Cache-Control: private, no-store.
• Quotes expire, paid submission is idempotent, and server-side account limits still apply.
• The assistant cannot claim that generated content is automatically cleared for publication or commercial use.

Troubleshooting

The browser returns to 127.0.0.1 and cannot connect

The desktop client's temporary callback listener is no longer running or used a different port. Keep the client open, restart the connection from that client, then approve the new request. Do not reuse an older authorization URL.

Authentication is requested again

The OAuth grant may have expired or been revoked. Reconnect and complete a fresh browser approval. Never paste a bearer token into the MCP address.

HTTP 401 / JSON-RPC -32001

The request has no valid bearer grant. Let the client follow the WWW-Authenticate challenge and its resource_metadata discovery URL.

INTERNAL_ERROR

Keep the returned correlationId, stop repeated submissions, and contact MaxVideoAI support if the failure persists.

A mode is missing

Ask for get_model_details again. A model can support several modes through different provider routes; one unavailable route must not be interpreted as the whole model being disabled.

Current boundaries

This is not a general-purpose REST API and does not issue customer API keys. It does not control the Studio timeline, edit arbitrary source video, manage customer callback endpoints, collect payment details inside chat, expose hidden models, or automatically approve media for publication.

Docs overview · Content safety checks · Getting started with MaxVideo AI