Which should I use to find the code causing a production exception?

Sentry. search_issues finds the grouped error, search_issue_events drills into individual occurrences, get_issue_tag_values scopes affected releases and environments, and analyze_issue_with_seer returns a root-cause analysis with a concrete code fix. Grafana's tools work at the metrics and logs layer, not the exception and stack-trace layer.

Is either server available as a hosted OAuth endpoint?

Sentry is. Its canonical deployment is the hosted remote endpoint at https://mcp.sentry.dev/mcp with OAuth, and it works against both Sentry SaaS and self-hosted instances. Grafana's server runs locally over stdio via Docker with a service account token; it does not offer a hosted OAuth endpoint.

Grafana vs Sentry

Grafana MCP and Sentry MCP are both monitoring-observability servers, but they address different problems in the ops loop. Grafana is a visualization and metrics layer: the agent can search and read dashboards (search_dashboards, get_dashboard_by_uid), run PromQL queries against Prometheus (query_prometheus, query_prometheus_histogram), run LogQL queries against Loki (query_loki_logs, query_loki_patterns), fetch Pyroscope profiles (fetch_pyroscope_profile), manage alert rules and OnCall schedules, and drive Grafana Incident with Sift automated investigations (list_sift_investigations, find_error_pattern_logs, find_slow_requests). Sentry is an error-tracking layer: the agent searches grouped issues and individual events (search_issues, search_events, search_issue_events), reads stack traces, inspects tag distributions (get_issue_tag_values), retrieves replays (get_replay_details) and profiles (get_profile_details), and invokes Seer (analyze_issue_with_seer) for AI-powered root-cause analysis with concrete code fixes.

How they compare

Dimension	Grafana	Sentry
Primary observability job	Dashboards, metrics, and logs: the agent queries time-series data with PromQL and LogQL, reads panels, and correlates signals across datasources to understand system behavior over time.	Error tracking and debugging: the agent reads grouped issues, drills into individual event stack traces, inspects tag distributions, and runs Seer root-cause analysis to find the specific code responsible for a production error.
Tool surface	Large and broad: dashboards, datasource listing, PromQL queries (query_prometheus, query_prometheus_histogram, list_prometheus_metric_names), LogQL queries (query_loki_logs, query_loki_stats, query_loki_patterns), Pyroscope profiles, Sift investigations, Grafana Incident, OnCall schedules and shifts, alert rules, and annotations.	Focused on the issue lifecycle: search_issues, search_events, search_issue_events, get_issue_tag_values, update_issue, analyze_issue_with_seer, get_replay_details, get_profile_details, find_organizations, find_projects, find_releases, create_project, create_team, create_dsn, and docs search.
AI-assisted root cause	Sift investigations (list_sift_investigations, get_sift_investigation, get_sift_analysis, find_error_pattern_logs, find_slow_requests) automate pattern detection across logs and traces, but output is investigative data rather than a code fix.	analyze_issue_with_seer sends a production error to Seer, Sentry's AI agent, which returns a root-cause analysis with specific code fixes ready to apply. use_sentry provides a natural-language interface to the full Sentry surface.
Incident and on-call management	Grafana Incident (list_incidents, create_incident, add_activity_to_incident) and OnCall (list_oncall_schedules, get_current_oncall_users, list_oncall_teams, list_alert_groups) are first-class tools for managing an active incident and notifying responders.	No dedicated incident-management or on-call tools. The agent can update_issue status and assignment, and search across projects to triage, but on-call scheduling lives outside Sentry's scope.
Deployment and auth	Local stdio via Docker (grafana/mcp-grafana with -t stdio flag), authenticated with a GRAFANA_SERVICE_ACCOUNT_TOKEN against a GRAFANA_URL. Works with self-managed Grafana 9.0+ and Grafana Cloud. No hosted OAuth endpoint.	Hosted remote endpoint at https://mcp.sentry.dev/mcp with OAuth (works against Sentry SaaS and self-hosted instances), plus a local stdio package via npx authenticated with SENTRY_ACCESS_TOKEN.

Verdict

Grafana MCP and Sentry MCP complement each other more than they compete. Grafana owns the visualization and metrics layer: use it when the task involves PromQL or LogQL queries, dashboard inspection, Pyroscope profiling, OnCall scheduling, or driving Grafana Incident through a Sift investigation. Sentry owns the error-tracking layer: use it when the task is finding which code threw an exception, reading a stack trace, inspecting tag distributions to scope affected releases, or running analyze_issue_with_seer for an AI-generated fix. Teams that run both can use Grafana to detect anomalies in time-series data and Sentry to pinpoint the source-code cause.

FAQ

Which should I use to find the code causing a production exception?: Sentry. search_issues finds the grouped error, search_issue_events drills into individual occurrences, get_issue_tag_values scopes affected releases and environments, and analyze_issue_with_seer returns a root-cause analysis with a concrete code fix. Grafana's tools work at the metrics and logs layer, not the exception and stack-trace layer.
Is either server available as a hosted OAuth endpoint?: Sentry is. Its canonical deployment is the hosted remote endpoint at https://mcp.sentry.dev/mcp with OAuth, and it works against both Sentry SaaS and self-hosted instances. Grafana's server runs locally over stdio via Docker with a service account token; it does not offer a hosted OAuth endpoint.