Features
MCP
Agents read the same screens you do through the Model Context Protocol, and act on them with a write token.
POST /mcp speaks MCP over HTTP. Create a token on Settings, Tokens:
- A read token gets the read tools: logs, traces, tasks, issues, processes, metrics, the status page.
- A write token also gets the tools that change something: resolve an issue, pause an alert, requeue a task, declare an incident. Actions that reach your app run as the person who made the token.
Payloads never leave through MCP: tools return fields and summaries, with secrets already masked at ingest.
Connect an agent
.mcp.json
{
"mcpServers": {
"vigie": {
"type": "http",
"url": "https://vigie.acme.dev/mcp",
"headers": { "authorization": "Bearer vm_…" }
}
}
}
Start with get_context_pack: it returns what this Vigie watches, the services, the task kinds, the processes and your definitions, so the agent knows your app's words before it asks.
What it can do
More than a hundred tools, grouped like the screens:
| Area | Tools |
|---|---|
| Logs and traces | search_logs, list_logs, get_event, get_trace, get_trace_logs, list_traces, get_request |
| Tasks | list_tasks, get_task, get_task_timeline, list_task_failures, get_task_health, requeue_task, cancel_task, bulk_task_action |
| Issues | list_issues, get_issue, resolve_issue, mute_issues, create_linear_issue |
| Processes | list_processes, get_process, list_stuck_instances, get_process_instance |
| Alerts | list_alerts, create_alert, preview_alert, pause_alert, export_alert_as_code |
| Status page | get_status_page, declare_incident, post_status_update |
| Releases | declare_deployment, get_release_verdict |
| Data | run_query, list_tables, read_table |
Every page and every defineTool becomes a tool too.