General questions
Answers to the most common questions about Durin, MCP, and governed agent access.
Which guide should I start with?
If you are creating an organization, use the quickstart to complete onboarding and your first reviewed read. If you are joining a workspace, start with MCP client setup after your invitation and account link are complete. Administrators adding a provider should use Add a new connection.
You can read every knowledge-base article without an account. The public MCP simulator is also available before connecting company systems. Its illustrative outcomes help explain the flow; they are not evidence that your production provider is configured.
What is MCP?
MCP (Model Context Protocol) is a standard protocol for AI agents to discover and invoke tools. Durin acts as an MCP gateway — it intercepts MCP requests and applies governance before forwarding them to the upstream tool provider.
Do I need a Durin account to read the docs?
No. The public docs explain the MCP workflow, simulator, decisions, and administrator tasks before you create an evaluation workspace.
Which MCP clients are supported?
Use an MCP client compatible with Durin’s remote HTTP transport and authorization flow, such as a supported version of Claude Code, Cursor, or OpenAI Codex. Exact setup syntax, authorization, and compatibility depend on the client and version.
Does Durin store my data?
Durin records decision metadata (who asked, what tool, what policy decision, what outcome) for audit purposes. Metadata-only retention is the default. Payload inspection, retention, and export should be configured for your deployment.
Can I use Durin without connecting production systems?
Yes. The evaluation flow supports policy, approval, and audit review before company systems are connected. You can use the MCP simulator to see allowed, denied, and approval-required examples without signing in.
Does Durin work with private or on-premise tools?
Durin is a hosted SaaS. Private upstream reachability requires a supported mechanism or an approved customer-side bridge. Universal private reachability is not claimed.
How does Durin connect to GitHub, Jira, or Slack?
Durin routes governed requests through configured provider connections. Authentication depends on the provider, deployment, and configured credential method; complete the connection’s setup guide before activation. Durin never passes its own bearer token to an upstream provider.
Why can I sign in but still see no tools?
Durin sign-in establishes who you are. Tool access also depends on organization membership, the installation’s enabled state, team scope, personal provider authorization, and the reviewed tool and resource boundary. A correct sign-in does not satisfy those other checks automatically.
Confirm that you are using your own issued endpoint in the intended workspace. Ask an administrator to check the relevant installation and team scope, then complete your provider sign-in and connection test where required. Start with a reviewed read whose resource is known to be allowed.
Is an allowed decision the same as a successful request?
No. The decision describes authorization; the execution outcome describes what happened after that decision. A provider can fail or time out after Durin allows a request. Review both fields before deciding whether another attempt is appropriate.
An approval-required response means the action has not run. A different administrator must review the exact request, and the requester must retry it before the grant expires. Production writes remain blocked by the current execution boundary, so use the sandbox to learn this workflow.
Does one endpoint give me every provider’s permissions?
Your personal endpoint is one entry point for the reviewed connections available to your identity. It does not merge all provider accounts into one unrestricted credential. Each connection retains its authentication, team, tool, and resource checks.
When one provider works and another does not, investigate the failing connection’s access and setup status. Avoid changing a workspace-wide policy merely because a separate provider grant is missing.