AI & MCP¶
WitFoo Analytics includes an in-app AI assistant and a Model Context Protocol (MCP) connector that lets AI clients work with your appliance's data.
Components¶
Configuration¶
Configure AI providers, models, and API keys.
Chat Interface¶
Use the AI chat panel for security analysis assistance.
MCP Connector¶
Connect AI clients such as Claude Desktop, Claude Code and Cursor to your appliance.
Overview¶
The AI assistant helps analysts with:
- Incident investigation and triage guidance
- Artifact analysis and enrichment interpretation
- Report generation and summary
- Compliance mapping assistance
MCP¶
AI clients connect to the appliance over MCP. Every connection acts as a real user, with that user's live permissions, and an administrator can see and end any of them.
- Users connect with a personal MCP key they create under Profile > Security > MCP connections.
- Apps can sign in through the appliance's own authorization server. See App sign-in (OAuth 2.1).
- An optional relay, off until an operator turns it on, reaches the appliance's MCP endpoint without an inbound port.
- MCP is included with Analytics Pro, Analytics Max and Reporter Pro. See Licensing.
Connections by personal key, by app sign-in and over HTTP /mcp are off by default. See MCP Configuration and the tool reference.
Permissions¶
| Permission | Name | Access |
|---|---|---|
ai:read | Read AI | View AI features and read chat assistant sessions |
ai:write | Write AI | Send messages; start, rename and delete chat sessions |
ai:manage | Manage AI | Configure AI providers and their API keys, and open the AI Setup page |
ai:export | Export AI | Export chat history and conversations |
mcp:connect | Connect MCP | Create, rotate and revoke your own personal MCP keys, approve and revoke the apps that connect as you, and open an MCP session that runs with your own permissions |
mcp:manage | Manage MCP | Set the organization's MCP mode, list and revoke any MCP key in the organization, and manage app sign-in clients and connected apps |
- Every built-in role holds
ai:readandai:write, directly or through Administrator. - A user with Read AI but not Write AI can open their chat sessions but cannot send or change anything.
- No built-in role is seeded with
mcp:connect. Admin and Platform Admin hold it through Administrator, and any role that holdsmcp:manageholds it too. Grant it to anyone else through a custom role. - The AI Setup page opens for
ai:manage. Its MCP Authentication section is shown to holders ofmcp:manage, so an administrator who manages MCP in the UI needs both.
Custom roles and the assistant at upgrade
On the first start of 1.2.0, existing custom roles receive ai:read and ai:write once. Removing them afterwards sticks. Custom roles created later do not get them automatically.
Permission changes reach a user at their next sign-in, and the server may reuse a decision for up to 5 minutes after that. MCP connections read a user's roles on every request and see a change on the next one.
See Permissions for the full catalog.