OfflineCreator MCP security and privacy directory

Check the remote OAuth sequence

The live protected-resource document identifies https://mcp.offlinecreatorstudio.com/mcp as the resource, points clients to the same-origin authorization server, advertises models, read, and generate scopes, and lists Authorization-header bearer tokens. The authorization-server metadata advertises authorization-code and refresh-token grants, a revocation endpoint, and S256 PKCE.

Compare that metadata with the current MCP authorization requirements rather than assuming discovery alone proves a secure client flow. The July 28, 2026 specification requires clients to send the resource parameter and use PKCE, and requires servers to validate that tokens were issued for the intended MCP resource instead of passing an inbound token to an upstream API. OfflineCreator's public product page says OAuth connections can be revoked from Settings, while the live authorization metadata identifies the revocation endpoint; these public sources do not by themselves prove post-revocation token behavior.

Discover

Protected-resource and authorization-server metadataVerify the resource, issuer, endpoints, supported scopes, and S256 before consent.

Consent

models, read, or generateApprove only the capabilities the client needs.

Revoke

One OAuth connectionRevoke the OAuth connection from Settings; personal API keys are managed separately.

Record what was verified and what remains untested

The privacy policy says sampled API request logs can contain route, status code, key identifier, and a truncated one-way network-address hash, and sets a retention ceiling of up to 12 months for sampled security and audit logs. It separately says key secrets are not written to application logs. This page does not extend those statements into a claim that every log path excludes every prompt or media field.

Historical verification notes are not used as supported public evidence on this page. The supported claim is limited to the product page's documentation of the recommended remote OAuth endpoint. The separately retrieved metadata establishes advertised protocol fields, not a deployment label, external certification, or guarantee of current behavior.

Evidence level

First-party policy, live metadata, and published package implementationNo independent audit or penetration-test report was located in this article.

Product-fact freshness

Verified August 8, 2026Monthly refresh required; claims expire September 7, 2026.

Community signal

InsufficientNo community claim is used because the engine returned no usable evidence.

Apply least privilege by capability

Treat the three advertised scopes as separate grants. In the published package implementation, models maps to list_models; read maps to get_credits, list_generations, get_generation, wait_generation, and download_output; and generate maps to generate, upload_input, and cancel_generation. When the server receives a nonempty scope list, it omits a tool whose required scope is absent. When scopes are missing or empty, the published implementation registers all tools, so callers must not treat an absent scope list as least privilege.

The current MCP specification requires resource-bound authorization. Confirm the consent screen, the granted scope set, and the tools visible after connection; metadata describes what can be requested, while the post-consent tool list shows what the client actually received.

models

Model and published-cost discoveryUse for catalog inspection without account-history or generation access.

read

Balance and generation statusAdd when the client must inspect account state or existing jobs.

generate

Create and manage media jobsReserve for clients authorized to spend credits and handle media.

Choose the transport without moving secrets into unsafe places

OfflineCreator recommends the hosted OAuth endpoint for remote clients. The current product page keeps local stdio and a legacy remote API-key path as fallbacks through @offlinecreator/mcp. Remote OAuth avoids placing a personal API key in the client configuration; fallback keys still belong in secret configuration, never in a URL or source control.

For personal API keys, the current privacy policy says the secret is shown once and stored only as a one-way hash alongside a display prefix, scopes, and last-used timestamps. It also says users can revoke one key or all keys from Settings. The public sources reviewed here do not expose the hashing construction or prove how malformed stored scopes are handled, so this page makes no stronger implementation claim.

Preferred

Remote OAuth over HTTPSBrowser consent and revocable grants; no personal API key in the client file.

Fallback

Local stdio with a personal API keyKeep the key in environment or secret configuration and restrict its scopes.

Legacy

Legacy remote API-key fallbackUse only as the documented legacy remote API-key fallback.

Continue with the boundary you need to inspect

Move from this directory to the narrower page for prompt processing, media access, or personal API-key handling. Return to the Privacy Policy for retention, provider, analytics, and rights disclosures.

What this article covers

This article explains the top-level OfflineCreator MCP security directory. Keep detailed prompt retention, media access, upload jailing, audit-log fields, key internals, and incident procedures on their narrower pages. Consolidate this directory into /privacy if current security documentation, policy, and implementation evidence can no longer support a distinct reviewer route.

This article does not treat community or social retrieval as proof of OfflineCreator quality, pricing, or customer results. Product statements stay limited to documented Studio routing, quotes, and provider disclosure for offlinecreator mcp security.

Coverage remains degraded. Reddit returned seven records before HTTP 429, Instagram failed with HTTP 404, and X was unavailable because optional credentials were not configured. All 11 YouTube records lacked publication dates in the raw export and reported zero captured views, likes, and comments. Generic security tutorials and promotional audit checklists were not treated as current product verification.

Community security reports are useful threat prompts, not proof of OfflineCreator behavior. Two records carried unsafe or insufficiently bounded security guidance, and three could not be verified from their available context. The privacy policy, live OAuth metadata, version-pinned package implementation, and current MCP authorization specification remain the stronger evidence. This pass did not perform a penetration test, external audit, post-revocation token test, or fresh OAuth acceptance run.