Coroot

Coroot MCP turns your Coroot observability data into a set of well-typed tools that an AI assistant can call to perform root-cause analysis. It exposes incidents, health summaries, and related context over an MCP HTTP endpoint, enabling your AI agents and tools to triage, summarize, and investigate issues without leaving your editor or IDE.

How to use

You connect an MCP-aware AI assistant to the server’s HTTP endpoint at /mcp to access a curated set of read-only tools. These tools provide incident lists, incident details, root-cause analysis, health summaries, traces, logs, and more. Point your MCP client at the endpoint and start calling tools like list_recent_incidents, get_incident_details, summarize_incident_root_cause, and investigate_incident. Tools return compact, machine-friendly payloads that your downstream dashboards and postmortems can consume.

How to install

Prerequisites: you need Java 21+ and Maven 3.9+ installed on your system. You also need access to a Coroot instance for production use and an OpenAI-compatible API key for Spring AI.

# Quick start in stub mode (no Coroot required)
./mvnw spring-boot:run -Dspring-boot.run.profiles=stub-coroot

# Quick start against a real Coroot instance requires environment setup described later

Configuration and runtime prerequisites

Configure essential environment variables in your runtime environment. The following are the variables used by the server and its MCP integration.

OPENAI_API_KEY=sk-...
COROOT_API_URL=https://coroot.your-company.com
COROOT_DEFAULT_PROJECT_ID=production
MCP_AUTH_TOKEN=optional-bearer-token
# Optional local test key, if you want to connect using a stub profile
OPENAI_MODEL=gpt-4.1-mini

Using with a real Coroot instance

To run against a real Coroot, start the server with the required environment variables set, then query the MCP endpoint at /mcp.

export OPENAI_API_KEY=sk-...
export MCP_AUTH_TOKEN=optional-token
export COROOT_API_URL=https://coroot.your-company.com
export COROOT_DEFAULT_PROJECT_ID=production

./mvnw spring-boot:run

MCP endpoint for clients

The MCP endpoint is available at the following HTTP URL. Use a POST request with the appropriate JSON-RPC 2.0 payload to call tools.

POST http://localhost:8080/mcp

Security and access

If you enable an authorization token, include it in requests to the MCP endpoint as a Bearer token. The supported token setting is MCP_AUTH_TOKEN and, when configured, clients must send Authorization: Bearer <token>.

Notes

The server is stateless and relies on external backing services (Coroot and the AI provider) for data and analytics. Tools are read-only and designed to be safe for exposure to assistants.

Available tools

list_projects

Returns the Coroot projects the current API key can access.

list_recent_incidents

Lists recent incidents with id, title, service, severity, and timestamps, with optional filters.

get_incident_details

Returns detailed incident context (summary, suspected root cause, affected services, metrics, timeline).

summarize_incident_root_cause

Generates a natural-language explanation of an incident’s likely root cause, blast radius, and remediation steps.

investigate_incident

Performs a full incident investigation and returns an RCA summary plus a structured JSON payload.

get_applications_overview

Returns an overview of application health for a project (per-service status and key indicators).

get_service_health

Returns a compact health snapshot for a single service.

get_nodes_overview

Overview of node health for a project (CPU, memory, network, disk, IPs).

get_deployments_overview

Returns recent deployments with service, version, status, age, and event summaries.

get_risks_overview

Returns a summary of risks for a project (exposure, availability).

get_traces_overview

Returns a tracing overview for a project, optionally filtered by a query.

get_application_traces

Returns recent spans for a single application over a time window, summarized for analysis.

get_logs_overview

Returns a logs overview for a project, optionally filtered by a log query.

get_application_logs

Returns recent log entries for a single application with severity breakdown.

get_costs_overview

Returns a cost overview for nodes and applications in a project.

get_slo_overview

Returns SLO availability and latency objectives at project and per-service level.