Most observability connections stop at read access. This one closes the loop: the agent investigates in APL, then creates the dashboard chart and the monitor in the same conversation. Query cost limits are the guardrail that makes letting it run unattended reasonable.
Axiom hosts this server at https://mcp.axiom.co/mcp and also ships it as a binary you can run yourself. Either way an agent gets your datasets, the APL query engine, and the dashboard, monitor and notifier objects around them.
- queryApl and queryMetrics run queries against your datasets; listDatasets and getDatasetSchema tell the agent what is there
- Dashboards created, updated, exported and deleted — updateDashboardChart patches a single chart by its ID instead of rewriting the whole document
- Monitors and notifiers created, updated and deleted, with checkMonitors for current status and getMonitorHistory for past alerts
- Metric discovery through listMetrics, listMetricTags, getMetricTagValues and searchMetrics
- Prebuilt prompts for anomaly detection, data quality investigation, performance baselines, cross-dataset correlation, monitor health and exploring a dataset nobody knows
An Axiom account. The hosted endpoint uses OAuth and agents never see your token; an agent that cannot do browser OAuth sends a personal access token plus the organization ID as headers. Running it yourself uses an API token instead, and the prebuilt prompts are not available in that mode. Hosted query results route through US infrastructure.
One command plus a key — go install github.com/axiomhq/axiom-mcp@latest, then supply credentials