Zabbix AI CLI — Zabbix MCP server and CLI for AI agents

zabbix-ai-cli is a Zabbix MCP server and command-line client in one Go binary. It gives Claude Code, Claude Desktop, Codex, Cursor and any other Model Context Protocol client task-shaped access to Zabbix — what is broken right now, why a host is silent, why an alert never arrived — with bounded output, a stable JSON contract, and no change to Zabbix without a person's approval.

It is for the people who get paged: SRE and DevOps teams running Zabbix who want an agent to triage an incident without handing it the whole API.

Requires Zabbix 6.4 or newer. Bearer-token authentication arrived in 6.4 and is the only scheme implemented; earlier versions expect the token in the request body instead.

It never contacts a language model. The AI decides, this program executes, Zabbix monitors.

zabbix-ai-cli login
zabbix-ai-cli problems list
zabbix-ai-cli host investigate server01
zabbix-ai-cli alert why 757474
zabbix-ai-cli mcp

Why not another Zabbix API wrapper

Wrapping host.get, problem.get and history.get hands the agent the API's sharp edges along with its power. On a current Zabbix 7.4 server:

• problem.get has no selectHosts, so a problem list arrives without hosts.

• history.get defaults to the numeric-unsigned table and returns nothing for a float item — silently, with no error.

• item.lastvalue and item.lastclock still exist and have returned a constant "0" for several major releases.

• host.available was removed in 5.4; availability lives on the interface.

• event.acknowledge takes a bitmask whose own documentation contradicts itself.

• searchWildcardsEnabled: true disables implicit substring matching, turning a name fragment into an exact match that quietly finds nothing.

Every one of those produces a confident, wrong answer rather than an error. This tool absorbs them behind commands that describe the task instead of the endpoint.

Related MCP server: Zabbix MCP Server

What it does

Safety

Read operations run immediately. Nothing else does.

There is no MCP parameter that applies a change, and adding one would not help: a confirmation an agent can send is a confirmation prompt injection can send. The approval lives at a terminal, outside the model's context.

$ zabbix-ai-cli maintenance create "ms*" --for 2h

PLAN pl_cc89d2e87d15

Create maintenance "ms* (2h0m)" for 2h0m, 2026-08-21T05:39:00Z to 2026-08-21T07:39:00Z

Affects:
  host ms1.8qw.ru
  host ms10.8qw.ru
  ...

Risk: write
Expires: 2026-08-21T05:54:22Z

Nothing has changed yet.
To apply it: zabbix-ai-cli approve pl_cc89d2e87d15

Before a plan runs, its parameters are re-hashed, its deadline checked and its preconditions re-read from Zabbix. A window that has been replaced since the plan was made is refused, not deleted. Every applied change is appended to an audit log.

This matters more than a refusal would. When the tool this replaces blocked a write, the work was done anyway with a token copied out of a container — losing the audit trail without preventing anything. A permitted path that is recorded beats a refusal that gets routed around.

Install

Prebuilt archives and the ghcr.io image are published with each tagged release. Until the first tag lands, build from source with either method below.

For a host-native binary, use Go 1.25 or newer:

go install github.com/stufently/zabbix-ai-cli/cmd/zabbix-ai-cli@latest

# Or build the current checkout.
mkdir -p bin
go build -trimpath -o bin/zabbix-ai-cli ./cmd/zabbix-ai-cli

The Make targets are container-first and do not require Go on the host:

make build      # Linux binary in ./bin, built inside Docker
make docker     # Linux container image for the MCP server

Configure

zabbix-ai-cli login --profile prod

It asks for the URL and the API token, verifies the token against the server, and stores it. The token is never accepted as a flag, because flag values are visible in shell history and in the process list; pipe it in instead:

printf %s "$TOKEN" | zabbix-ai-cli login --profile prod --url https://zabbix.example.com --token-stdin

Writes are off until a profile is granted the scope for them:

zabbix-ai-cli profile scopes prod --add maintenance

See docs/authentication.md for the resolution order and the headless and container cases.

Add the Zabbix MCP server to your AI client

The MCP client never sees the Zabbix token. It is resolved inside the server process from the profile you configured, so the credential never enters a model's context or a client's configuration file.

Claude Code

claude mcp add zabbix -- zabbix-ai-cli mcp --profile prod
zabbix-ai-cli skills install claude

Claude Desktop

claude_desktop_config.json:

{
  "mcpServers": {
    "zabbix": {
      "command": "zabbix-ai-cli",
      "args": ["mcp", "--profile", "prod"]
    }
  }
}

Codex

# ~/.codex/config.toml
[mcp_servers.zabbix]
command = "zabbix-ai-cli"
args = ["mcp", "--profile", "prod"]

zabbix-ai-cli skills install codex

Cursor, Windsurf, VS Code and other MCP clients

Any client that speaks stdio takes the same two fields — command zabbix-ai-cli, arguments ["mcp", "--profile", "prod"]. For a client that wants HTTP instead:

zabbix-ai-cli mcp --http 127.0.0.1:8000

It refuses a routable address unless you pass --allow-remote together with a bearer token, because an unauthenticated MCP endpoint is an unauthenticated route into Zabbix. See docs/mcp.md.

Docker

docker run --rm -i \
  -e ZABBIX_AI_CLI_URL=https://zabbix.example.com \
  -e ZABBIX_AI_CLI_TOKEN_FILE=/run/secrets/zabbix \
  -v /path/to/token:/run/secrets/zabbix:ro \
  ghcr.io/stufently/zabbix-ai-cli:latest mcp

MCP tools

Fourteen tools, not two hundred. A large tool surface costs an agent context before it has done anything, and most of it is never called.

zabbix_problems           zabbix_metrics_latest      zabbix_unreachable
zabbix_problem            zabbix_metrics_history     zabbix_maintenance_list
zabbix_hosts              zabbix_alert_why           zabbix_api_call
zabbix_host_status        zabbix_resolve             zabbix_plan_create
zabbix_host_investigate                              zabbix_plan_status

Write operations do not get one tool each. zabbix_plan_create takes an operation enum generated from the same registry the CLI is built from, so the tool surface does not grow as operations are added.

JSON contract

{
  "ok": true,
  "data": {},
  "warnings": [],
  "meta": {
    "returned": 50,
    "total": 381,
    "truncated": true,
    "truncated_reason": "row_limit",
    "partial": false,
    "zabbix_version": "7.4.10"
  }
}

Errors carry a stable code, whether retrying is worthwhile, and what to do next:

{
  "ok": false,
  "error": {
    "code": "AUTHENTICATION_FAILED",
    "message": "Zabbix rejected the configured API token",
    "retryable": false,
    "suggestion": "run 'zabbix-ai-cli login' to configure a new token"
  }
}

zabbix-ai-cli schema prints every operation, its parameters and its JSON Schema, so an agent can learn the tool programmatically instead of guessing at flags.

Full details in docs/json-output.md.

Documentation

• Authentication and profiles

• Command line

• MCP server

• Skills

• JSON output and exit codes

• Security model

• Architecture and design decisions

FAQ

What is a Zabbix MCP server?

An MCP server is a small program that exposes a system to an AI client over the Model Context Protocol. A Zabbix MCP server lets Claude, Codex, Cursor and similar clients query Zabbix — problems, hosts, items, events, maintenance — as tools, instead of the model guessing at curl calls against the JSON-RPC API.

Can an AI agent change my Zabbix through this?

Not on its own. Read operations run immediately. Every write produces a plan and stops. Applying that plan is a command a person runs in their own terminal: zabbix-ai-cli approve <plan-id>. There is no MCP parameter that applies anything, and a test fails the build if one is ever added.

Does it send my monitoring data to an AI provider?

No. This program never contacts a language model. It talks to Zabbix and prints JSON. Whatever your MCP client does with that output is between you and your client.

Which Zabbix versions are supported?

Zabbix 6.4 and newer, because bearer-token authentication arrived in 6.4. Developed and tested against Zabbix 7.4.

Do I need Go installed?

No. make build compiles inside Docker and needs nothing on the host but Docker itself. go install is there for a host-native binary, and each tagged release publishes archives for Linux, macOS and Windows with checksums, along with a container image on ghcr.io.

How is this different from an MCP server that wraps the Zabbix API?

A thin wrapper hands the agent the API's sharp edges: problem.get without hosts, history.get silently returning nothing for float items, lastvalue frozen at "0". Those produce confident wrong answers rather than errors. This tool answers questions — "what is broken", "why did this alert not arrive" — and absorbs the traps behind them. It also ships fourteen tools rather than two hundred, because a large tool surface spends an agent's context before it does any work.

Can I still call the raw Zabbix API?

Yes, through api call, under the same approval rules. Methods that hand out credentials or execute code are refused outright — including the long way round, such as creating a script and having an action run it.

Does it work without an AI client at all?

Yes. It is a normal CLI with human-readable tables, JSON output and documented exit codes, so it is equally usable from a shell or a CI job.

Compatibility

Zabbix 6.4 and newer. Bearer-token authentication arrived in 6.4 and is the only scheme implemented. Version-dependent behaviour is asserted explicitly, so an incompatibility is reported rather than returning an empty result.

Developed and tested against Zabbix 7.4.

License

Apache-2.0. See LICENSE.

Zabbix is a trademark of Zabbix LLC. This project is an independent open-source project and is not affiliated with or endorsed by Zabbix LLC.