Opsin MCP Server
Overview
The Opsin MCP server connects Opsin to an MCP-compatible AI app. Opsin finds AI security risks in your organization: issues (risky agents, overshared data), alerts (risky AI interactions), AI agents, AI activity, identities, and integrations. Through the server, the app can read this data and update issues and alerts.
Requirements
- An Opsin account.
- MCP access turned on for your organization. An admin must do this. [TODO: where the admin turns on MCP access]
- An MCP-compatible AI app.
- A role that allows the tools you need.
Connect
- Ask an Opsin admin to turn on MCP access for your organization.
- In your AI app, add a new MCP server. The steps depend on the app.
- Enter the server URL:
https://api.opsinsecurity.com/api/mcp - Sign in with your Opsin account. The server uses OAuth.
- Pick one organization.
- Pick how long to stay connected: 1 hour, 1 day, 7 days, or 30 days.
Tools
| Tool | What it does | Type |
|---|---|---|
| list_issues | Lists issues. | Read |
| get_issue | Gets one issue. | Read |
| list_alerts | Lists alerts. | Read |
| get_alert | Gets one alert. | Read |
| list_agents | Lists AI agents. | Read |
| get_agent | Gets one AI agent. | Read |
| list_ai_activity | Lists AI activity. | Read |
| get_interaction | Gets one AI interaction. | Read |
| list_identities | Lists identities. | Read |
| list_integrations | Lists integrations. | Read |
| list_data_categories | Lists data categories. | Read |
| list_assessments | Lists assessments. | Read |
| update_issue_status | Changes the status of an issue. | Write |
| update_alert_status | Changes the status of an alert. | Write |
| comment_on_issue | Adds a comment to an issue. | Write |
| comment_on_alert | Adds a comment to an alert. | Write |
| submit_feedback_to_opsin_team | Sends a short note to the Opsin team, not to your organization. | Write |
Permissions and data handling
- Every call runs as the signed-in user, in the chosen organization, with that user's permissions.
- A tool only appears if the user's role allows it.
- Reading the prompt text of an AI interaction needs permission to read chats.
- Comments are visible to everyone in the organization who can see the issue or alert.
- Tool results can contain customer data. The server does not store or train on tool results.
- You can end a connection at any time in Opsin under Settings > Integrations.
Limits
- Lists return at most 100 rows per call. The default is 25.
- AI activity covers the last 24 hours by default. The maximum is 30 days.
Support and contact
Email support@opsinsecurity.com.
