Shared context. Clear access.

Workspace access is shared.

Every active MCP key can read all clients, brands, and records in its workspace. Selecting a brand for a task is a retrieval choice, not an access restriction. Do not issue a workspace key to an external party on the assumption that it can only see one brand.

Owners and admins use the app to manage memory and agent access. Other teammates can work through MCP keys issued to them. Create a separate named key for each person or connection and revoke it when it is no longer needed.

Choose which connections can change memory.

Solo mode uses one workspace-level MCP writes switch. Agency mode uses a write setting on each MCP key. The server checks the current setting when authenticating requests; changes take effect on the key’s next request.

A writable key can apply memory changes directly. There is no pending approval queue. A Playbook can describe a review process, but its instructions do not replace write permissions. Keys cannot grant themselves additional authority or manage workspace memberships and access settings.

Treat MCP keys as credentials.

The full key is shown at creation. The saved key record contains a hash and an identifying prefix, not a recoverable copy of the original secret. Use private agent settings or an appropriate secret store, and keep keys out of shared documents, source control, screenshots, and support messages.

Revoke a compromised or unused key from Agent access. The service refuses requests from revoked, expired, or unknown keys. Revocation stops future requests; it cannot retract information an agent has already received.

History helps you trace changes.

Versioned records preserve earlier states, and audit records identify changes and their outcomes. Content can carry source and verification information. A writable connection and a verified fact are different things: access to write does not make an agent’s output correct.

Archiving preserves the record’s history. It should not be treated as permanent deletion. Contact support@agencymemory.ai for a privacy or deletion request.

Brand files use temporary links.

Where file storage is enabled, brand files use private storage and temporary signed links for authorized access. A download link can be used by anyone who holds it while it remains valid, so share it only with intended recipients. File availability depends on the service’s storage configuration.

Connected agents and tools have their own responsibilities.

When an agent retrieves context, that information enters the agent’s environment. Review the AI provider’s terms, retention settings, and suitability for your clients’ information before connecting it. Agency Memory does not control a third-party provider’s handling of retrieved content.

Platforms records account and asset relationships. It does not authenticate with Google Ads, Meta Ads, analytics providers, or creative tools. Access to those tools and authority to act remain separate.

A practical starting setup

• Give access only to people and agents trusted with the entire workspace.
• Use read-only access for the first retrieval and drafting workflow.
• Maintain approved facts and flag information that needs confirmation.
• Keep provider passwords and API secrets outside brand knowledge.
• Review active keys and remove connections you no longer use.

For the setup sequence, follow Getting started. For questions about information handling, read the privacy policy draft or contact us.