The analysis tools deduplicate and cluster before anything reaches the model, so a noisy window comes back as candidate causes rather than raw lines. The same server also builds pipelines, and it holds back on the irreversible parts — a pipeline created from a manifest stays a draft unless publish is explicitly set to true, and deletion is documented as needing your confirmation first.
Mezmo's remote server at https://mcp.mezmo.com/mcp, connecting an AI assistant or IDE chat agent to the Mezmo observability platform for root-cause analysis, log queries and pipeline management.
- Root-cause analysis over a relative or absolute window — logs fetched, deduplicated, clustered and analysed in one call
- Log shaping: group_logs_by_field for distributions like errors per app or p95 latency per host, get_log_histogram for volume over time, list_log_fields to find what you can filter on
- A correlated timeline that groups logs from many sources by app, host or any custom field
- OpenTelemetry trace work: the service graph, failed traces, spans for a trace_id, and find_degraded_service_edges for calls slower than a baseline period
- Pipelines listed, inspected, created from a manifest, edited node by node, paused and published — plus tap_pipeline_component for a live sample of events
- A query syntax with fielded search on app, host, level, namespace and pod, dotted paths for nested JSON, and automatic prefix matching
A Mezmo Service Key, generated in your dashboard under Settings > API Keys and sent as an Authorization: Bearer header. Clients that speak remote URLs point straight at the endpoint; stdio-only clients bridge through mcp-remote, which needs Node.js 18 or later.
One command plus a key — AUTH_HEADER="Bearer <SERVICE KEY>" npx mcp-remote https://mcp.mezmo.com/mcp --header "Authorization:${AUTH_HEADER}", then supply credentials