Security and Data Handling
Understand the boundaries of external-agent access and how data is handled.
On this page
Review statusWhat the MCP connection readsIdentity, revocation, and isolationAuditYour agent providerReporting a vulnerabilityReview status #
This page is a draft awaiting independent security review and an approved public vulnerability-reporting contact. It is not a production security assurance.
What the MCP connection reads #
The reviewed MCP contract exposes read-only access to one workspace's identity, active Team Members, Auras™, and Session History as the credential's role permits. A session read includes available summary, participants, report, and a bounded transcript window. Curation interviews are excluded.
There are no MCP write tools in this contract. This does not describe the permissions of other Marketplace applications.
Identity, revocation, and isolation #
Studio resolves the credential's workspace and role for each call. API keys have required expiry, can be revoked, and stop working when their creator no longer meets the role and membership requirements. A key's role cannot exceed its creator's.
A record in another workspace is not distinguishable from a nonexistent record in the MCP response. The control and its limitations are reviewed against SECURITY.md §1.33, §1.60, and §6; this page must not promise a broader platform guarantee.
Audit #
The gateway records a read before performing it. The reviewed contract records the tool, credential identity, request identifier, argument digest, and client IP. It does not establish a public retention commitment.
Your agent provider #
Once a client reads session data, that data is in the external provider's context. The client you choose and its provider's handling policies matter. Request only the data needed for the task.
Reporting a vulnerability #
The publication owner must supply and verify a monitored reporting channel before this page is approved. Do not include credentials or customer records in a public issue.