Skip to main content

    A data layer for AI SRE

    Give your AI agent live access to your production cluster. Topology, service dependencies, and the full observability stack (metrics, traces, logs, events, profiles) through a single structured surface.

    Live application topology

    The actual service map of your production cluster: what calls what, with what protocol.

    Historical performance

    Hours, days, or weeks of metrics, logs, traces, events, and profiles to compare 'now' against.

    All telemetry signals

    Metrics, traces, logs, events, and profiles. Exposed through a single, structured surface.

    What your agent gets access to

    The same model the Coroot UI uses, exposed as structured tools instead of dashboards. Read-only and powered by Coroot's RBAC, so each agent sees exactly what its user is allowed to see.

    Application topology & dependencies

    Live service map of your cluster: what calls what, over which protocol (HTTP, gRPC, SQL, Kafka, Redis), with full SLIs (latency, request rate, failed requests) captured by eBPF, no blind spots.

    Metrics

    Ad-hoc PromQL plus per-app RPS, error rate, latency percentiles, CPU, memory, and per-host utilization.

    Distributed traces

    Per-endpoint p50/p95/p99, error reasons with sample trace ids, slow-tail flamegraphs, and full trace-by-id.

    Logs

    App-scoped or project-wide log search by severity, text, time range, or log pattern.

    Events

    Alerts, SLO incidents, anomalies, deploys and rollouts, plus DB schema and config changes across the cluster, with the historical timeline.

    Profiles

    CPU and memory profiles, so the agent can ask why a service slowed down, not just observe that it did.

    Connect your agent in one command

    Standards-compliant Model Context Protocol over HTTP with OAuth 2.0. The agent runs with the user's Coroot RBAC permissions on every call.

    Claude Code
    claude mcp add --transport http coroot https://<your-coroot>/mcp
    Codex
    codex mcp add coroot --url https://<your-coroot>/mcp

    Full setup, auth flow, and the complete tool reference in the documentation.

    Make your agent investigate

    Coroot Enterprise Edition adds anomaly detection and root-cause investigation. Coroot's deterministic RCA does the heavy lifting before the LLM, so each investigation usually costs under $0.10 in tokens.

    Find anomalies across the fleet

    The agent gets a single call that returns every app currently in trouble: SLO violations and even sub-SLO error or latency spikes that the alerting rules have not caught yet.

    Run a full root-cause investigation

    Point the agent at one app and ask why it broke. Coroot walks the dependency graph, checks every candidate cause, and returns the root cause, the immediate fix, and a propagation map showing how the failure spread.

    Close the feedback loop on AI-written code

    Your agent ships code to production. Coroot tells it what happens next.

    How the new version performs

    RPS, error rate, and latency percentiles for the freshly deployed pods, side by side with the previous version. The agent can tell whether the change is a regression.

    New errors in logs

    Coroot's log-pattern detection groups stack traces and surfaces patterns that started appearing only after the deploy, so the agent finds new failure modes without reading every line.

    Hot code paths only on production

    CPU and memory profiles show which functions the new code actually spends time in at scale, including paths that never showed up in tests. The agent sees the real cost of what it shipped.

    Impact on databases

    Per-app database stats: connection counts, query rate, slow queries, locks, replication lag. The agent can tell whether the new version put unexpected pressure on Postgres, Redis, MongoDB, or Mysql.

    Connect your AI agent to your production

    Set up takes one command. The agent gets the full Coroot model: topology, SLIs, traces, logs, metrics, events, and profiles.

    ✓ Standard MCP over HTTP ✓ OAuth 2.0 ✓ Works with any MCP-compatible client