Usage policy

Review our usage policy to confirm your AI use adheres to our guidelines. If your implementation breaches these guidelines, we may ask you to remediate the breach or we may opt to remove it.

This policy applies to external partners building MCP-based experiences that integrate with the Canva MCP through Canva Connect APIs and tools.

It governs:

• Brand use and representation.
• UX and integration quality.
• Data privacy and security.
• Brand Kit and template handling.
• Agent permissions and execution behavior.
• API usage, rate limits, and credits.
• Distribution and review readiness.

Partners must comply with:

• Canva Terms presented during OAuth authorization.
• Developer Portal technical guidance.
• Platform review standards.

Canva reserves the right to suspend or withdraw MCP access for violations of this policy.

• Integrations must clearly indicate that users are accessing Canva MCP functionality.
• Use only compatibility language (e.g., “for Canva,” “works with Canva”).
• Don't imply the integration is built, endorsed, or certified by Canva unless formally authorized.

• Only approved Canva logo assets may be used.
• Logos must not be altered, recolored, distorted, or embedded within third-party wordmarks.
• Logo usage must not imply endorsement or certification.

Partners must:

• Define clear use cases.
• Map data flows explicitly.
• Implement secure OAuth flows.
• Validate inputs before MCP invocation.
• Monitor integration health.
• Implement robust error handling and retry logic.

Integrations must meet baseline accessibility standards:

• Alt text where applicable.
• Keyboard navigability.
• Adequate color contrast.
• Clear labeling.
• Plain language messaging.

• Integrations should align with User interface (UI) guidelines for Connect API.
• The experience should feel cohesive without mimicking native Canva features.
• Avoid parallel or duplicative flows that create user confusion.

Partners are responsible for:

• Securing their infrastructure.
• Respecting granted OAuth scopes.
• Protecting tokens and credentials.
• Handling webhooks securely.
• Monitoring and mitigating abuse.

Canva enforces:

• Platform-level protections.
• Health and error monitoring.
• Rate limiting.
• Integration suspension where necessary.

Canva-sourced data, prompts, outputs, Brand Kit elements, or design structures must not be used to train, fine-tune, evaluate, or improve external AI models.

• Only request scopes necessary for declared use cases.
• Don't request speculative or unused scopes.
• Authenticated flows must not be accessible without authentication.

When handling user content or Brand Kit data:

• Respect privacy preferences and opt-outs.
• Avoid unnecessary human review of generated content.
• Apply secure storage and transmission practices.
• Don't retain data beyond operational necessity.

Brand Kits retrieved through MCP:

• May only be used to generate or modify Canva designs.
• Must not be exported for non-Canva design generation workflows.
• Must not be extracted, cached, transformed, or reused outside Canva-rendered outputs.
• Must not be used to train, fine-tune, or enrich external AI systems.
• Must not be converted into generalized style embeddings.

MCP integrations must not crawl, pre-fetch, enumerate, or index user designs, folders, or Brand Kits without explicit user action.

• Canva templates must not be programmatically analyzed, decomposed, or replicated outside Canva’s design environment.
• MCP integrations must not expose template internals (layout logic, component hierarchy, token structure).

• When Brand Kit context is provided, outputs must use only the active Brand Kit.
• Brand assets from multiple Brand Kits may only be merged if the user has access to each Brand Kit and explicitly requests multi-brand output.

MCP agents must:

• Execute strictly within the authenticated user’s permission scope.
• Not escalate privileges beyond the OAuth token used at invocation.
• Not access designs, folders, or Brand Kits outside the user’s access graph.

For shared MCP agents:

• Shared agents must not execute using another user’s Canva OAuth token.
• Sharing an agent may only share its configuration and logic. Each user must authenticate their own Canva account before they can run the agent or access Canva-derived outputs.
• Each execution must evaluate permissions at runtime per invoking user.
• Clearly indicate what user content is being accessed or shared.
• Shared agents must not retain or reuse user context across users with different permissions.
• Cross-user caching of design metadata, Brand Kit data, or prompts is prohibited.
• Logs must segment user sessions logically.

• Agents may not act on behalf of another user without explicit delegation authorization.
• Delegated workflows must require explicit approval and scope validation.

• MCP tools follow per-user limits aligned with Connect API documentation.
• Some tools may have extended execution time (e.g., generate-design may take up to 60 seconds).
• OAuth endpoints enforce per-client protections and aren't partner-tunable.

Partners must:

• Implement exponential backoff on 429 responses.
• Avoid bulk automation that circumvents per-user limits.

• Integrations must monitor error rates.
• Persistent high error rates may result in suspension.
• Retry logic must be predictable and bounded.

• Canva’s credit model applies to applicable MCP tools (e.g., generate-design).
• Where possible, partners must show the user's Canva credit usage in the UX or tool call response.
• Integrations must not mask or obscure credit consumption.

Before launching a public MCP integration, partners must:

• Confirm compliance with Canva brand and UI guidelines.
• Validate naming and logo usage.
• Verify OAuth flows and scopes.
• Test rate-limit handling.
• Confirm secure data handling.
• Ensure error monitoring is in place.

• Review the Prohibited use for additional restrictions
• Overview of Canva Model Context Protocol (MCP) for AI assistants