Setting Permissions
Configure security policies for tools and resources.
Permissions control how Edison Watch governs AI tool usage. Correct classification is essential for organization security.
Configuration Location
Permissions are set per-server in the Servers page:
- Select a server from the sidebar.
- Click the Tools tab.
Security Flags
Each tool has four flags that determine its risk profile:
- Enabled: Can the tool be used at all?
- Write Operation: Does it send or modify data externally? (e.g., Slack, Email)
- Read Private Data: Does it access sensitive internal info? (e.g., Database, Files)
- Read Untrusted Data: Does it fetch external content? (e.g., Web Fetch)
The Lethal Trifecta
When a session combines Private Data + Untrusted Data + Write Operation, Edison Watch pauses the action for user approval.
ACL Levels
Access Control Levels (ACL) prevent high-sensitivity data from leaking to low-sensitivity destinations:
| Level | Meaning |
|---|---|
| PUBLIC | Non-sensitive data. |
| PRIVATE | Confidential or internal data. |
| SECRET | Highly sensitive data. |
Enforcement: AI cannot flow data from a higher ACL to a lower one (e.g., SECRET → PUBLIC).
Common Patterns
| Tool Type | Private | Untrusted | Write | ACL |
|---|---|---|---|---|
| File Read | ✓ | ✗ | ✗ | PRIVATE |
| Web Fetch | ✗ | ✓ | ✗ | SECRET |
| Messaging | ✗ | ✗ | ✓ | PRIVATE |
| Database | ✓ | ✗ | ✗ | SECRET |
Auto-Configuration
Edison Watch can suggest classifications using AI:
- Select a server.
- Click Auto-Configure.
- Review and adjust suggestions as needed.
Always review auto-classifications. Your specific security requirements may vary from the defaults.
Resources & Prompts
The same model applies to MCP resources and prompts via their respective tabs. Use the Enabled toggle to control access.
See the Security Model for a deep dive into the architecture.