From MCP security evidence to AI governance
Review the provider. Control the request. Keep the evidence.
The index informs provider review. Durin can apply access policies, independent human approval and audit records to supported MCP requests routed through its gateway.
Use an assessment to decide what to govern
A provider’s documented safeguards are one input to a connection review. Check the assessed configuration, current advisories and approval evidence. Then decide which identities may use the connection, which tools they need and which sensitive actions require review.
For requests routed through Durin, policy and current authority are checked before execution. Required approval comes from a separate reviewer, is bound to the exact request, expires and is single-use. Approval does not expand the provider’s permissions. Read Durin’s security controls and coverage boundaries.
Apply governance to the intelligence workflow
This is a deployment pattern for operators of an MCP security index. Put only named, narrowly scoped administration tools behind the governance boundary; retain the index’s own validation and publication authority.
| Operation | Governance pattern | Index responsibility |
|---|---|---|
| Collect evidence | Allow approved operators to request bounded collection tools. | Fetch configured hosts with size limits and timeouts; treat fetched text as data. |
| Review a provider | Scope review tools to the connection and resources an operator needs. | Resolve canonical identity and retain attributed source evidence. |
| Publish an advisory | Require independent approval for a specific proposed publication. | Validate evidence authority, optimistic revision and expected publication generation. |
| Change configuration | Restrict sensitive tools and review each requested change. | Preserve deterministic rubrics, signed administration and append-only history. |
| Investigate an incident | Inspect routed request evidence and revoke the relevant access. | Keep advisory revisions and distinguish a permitted action from its observed result. |
Durin records audit intent before a governed action and distinguishes a permitted action from a confirmed result. Coverage depends on requests using the governed route; bypass paths require separate network, endpoint and credential controls. Verify the deployment boundary.
How the Index protects its evidence today
- Scanning, scoring and ranking make no model calls. Versioned rules determine the result.
- Internal administration requires signed, scoped, short-lived operator assertions and replay protection.
- Advisory revisions and administration audit records are append-only.
- Provider evidence and scores require a verified browser session and bounded access.
- Detection feed manifests are signed, with generation, expiry and digest verification.
These controls are implemented by the Index itself. Adding Durin to the operator workflow is a separate integration, not an automatic consequence of listing a provider.
Index security rules · Assessment methodology · Durin security · Durin MCP setup
Questions about MCP security
What can Durin govern?
Durin applies identity checks, access policy, approval requirements and audit records to supported MCP requests sent through its gateway. Direct API calls, browser actions and local shell commands need their own controls.
Can Durin help protect an MCP security index?
Yes, as a deployment pattern: expose narrowly scoped operator tools through Durin, limit who can request them, require independent approval for sensitive publication or configuration changes, and retain the request and execution evidence. The index must still validate evidence and enforce its own publication rules.
Is this index already operated through Durin?
This guide describes how to add Durin to an intelligence workflow. It does not claim that the production Index operator path currently runs through the Durin gateway. The Index currently implements its own signed operator assertions, browser access checks and deterministic publication controls.
Can an AI agent change a provider score or publish an advisory?
Model output has no scoring or publishing authority in this index. Assessments follow versioned deterministic rules. Advisory publication requires resolved identity, attributed evidence and the index’s authorized publication path.
Start with a governed connection
Choose a supported MCP connection, set its access policy, test a sensitive request and inspect the resulting approval and audit evidence.