Hosted by ilert

The MCP servers ilert runs for you, and how ilert reaches systems that have no public endpoint.

ilert runs these MCP servers for you. You do not deploy or update them — you only create a credential for the system the server talks to, and attach it to an agent. Each server has its own page below with the exact permissions it needs and the steps to connect it.

How ilert reaches your system

Most servers connect straight to an endpoint you already expose over HTTPS. Four of them read from systems that usually have no public endpoint at all, so they go through a small component you deploy in your own network.

How it connects	Servers
Directly to your HTTPS endpoint	Elasticsearch, GitHub, Grafana, Graylog, InfluxDB, Netdata, Prometheus, ServiceNow
Through ilert-mcp-connector, deployed by you	ClickHouse, Kafka, MySQL, Redis
Through ilert-kube-agent, deployed in your cluster	Kubernetes

The connector and the kube agent exist so ilert never needs direct network access to a database, broker or cluster. You expose only the component, and it exposes only the read operations the MCP server uses.

One connector deployment serves ClickHouse, Kafka, MySQL and Redis together — deploy it once and enable the systems you need:

ilert MCP connector

Connect a server

The flow is the same for every server on this page:

1

Create the credential the server needs — an API token, a service account, a database user. The server's page lists the exact permissions

2

Navigate to Agents, open your agent, and go to the Tools tab

3

Find the server under ilert hosted MCP servers and click Connect

4

Enter the credential and, where required, the Host URL of your system

5

Enable the tools the agent may use, then save the agent

Connecting a server makes it available to that agent only. Other agents keep their own tool sets.

Available servers

Server	What the agent gets
Clickhouse	Databases, tables, schemas and server metrics (read-only)
Elasticsearch	Indices, log searches and aggregations (read-only)
GitHub	Pull requests and commit diffs (read-only)
Grafana	Loki logs, dashboards and Tempo traces (read-only)
Graylog	Streams and log searches (read-only)
InfluxDB	Measurements and time-series queries (read-only)
Kafka	Topics, consumer groups and consumer lag (read-only)
Kubernetes	Pod state, descriptions and logs — plus write tools: scale, patch CPU/memory, roll back, delete a pod
MySQL	Databases, tables, schemas and server metrics (read-only)
Netdata	Rooms, nodes, alerts and metric queries (read-only)
Prometheus	Metric queries and series metadata (read-only)
Redis	Keyspace, key metadata and server metrics (read-only)
ServiceNow	Records, incidents and cases — plus write tools: create, update and comment on incidents and cases

Only enable the tools an agent actually needs. The Kubernetes and ServiceNow servers include tools that change your systems, and every tool you enable is a tool the agent may decide to call.