Sardine MCP Server: Governed AI Access

MCP (Model Context Protocol) is the open standard that lets AI models and agents connect to external tools and data sources through a common interface, instead of a custom integration for every tool. Sardine is an MCP server that exposes a defined set of tools an AI agent can call. Barndoor sits at that connection point, applying access policies and audit logging to every tool call regardless of which AI client is making it.

Yes. Barndoor is SOC 2 Type II compliant. Barndoor enforces fine-grained access control, requires approval before destructive actions, protects sensitive data with data loss prevention, and logs every action for audit.

Barndoor tracks tool and schema changes on Sardine. New or changed tools inherit deny-by-default access until an admin reviews and approves them, keeping runtime enforcement current without manual reconfiguration. Every schema change is captured in the audit trail alongside who approved it.

Barndoor authenticates admins and policy managers through your OIDC or SAML 2.0 identity provider. Signing into Claude, Cursor, or ChatGPT still happens directly with those providers. Barndoor's integration point is the Sardine connection where access policies apply to what an agent can do once it's connected.

No. Barndoor brokers the OAuth connection to Sardine centrally, so users and agents never handle or store a standing token themselves. Access is granted and revoked through policy instead of shared credentials.

Barndoor gives each AI agent its own governed identity called an AgentProfile, separate from the human who configured or uses it. AgentProfile lets policies and audit logs attribute a given action to the specific agent, not a shared service token, so you can set different access rules per agent and revoke one agent's access without affecting others.

Yes. Barndoor supports customer-built MCPs, so internal tools and systems your team builds get the same policy enforcement and audit trail as Barndoor MCPs and official MCPs.

Yes. Barndoor's access control is fine-grained, down to the agent, team, group, user, MCP server, the specific tool call, and the data type involved. Every call is logged with a timestamp, the user or agent calling it, the client it came from, and the result of the policy applied, so you can see exactly who or what did what, when, and whether it was allowed, across any team or role.

Connect Sardine once, define your access policies, and go live across every AI client your team uses. Setup takes three steps: connect the server, define policies by role or agent, and roll out to Claude, Cursor, ChatGPT, and any other MCP-compatible client. Request a demo at barndoor.ai/get-a-demo to learn more.