How do I use zabbix-mcp?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@zabbix-mcp Show all active problems with severity High or above" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

zabbix-mcp

by lordraw77

Overview Schema Source Related Servers Discussions Admin

Python

Remote

zabbix-mcp

A Model Context Protocol (MCP) server that exposes Zabbix monitoring capabilities as callable tools for AI agents and MCP-compatible clients.

Features

30 tools across 11 categories (hosts, problems/triggers, items/history, maintenance, host groups, events, graphs, templates, inventory, actions, users, analytics, operational)
Transports: stdio (default) and Streamable HTTP — selectable via env var or CLI flag; the deprecated HTTP+SSE transport is still served for backwards compatibility
Protocol: MCP 2024-11-05 (negotiates up to 2025-06-18 over Streamable HTTP)
Auth: API token (Zabbix 5.4+) or user/password, loaded from .env
Bundled interactive CLI agent using NVIDIA, OpenRouter or Groq as LLM provider

Related MCP server: mcp-zabbix

Requirements

Python 3.11+
A Zabbix instance reachable from the server process (5.4+ recommended for token auth)
An API key for at least one LLM provider (if using the agent)
starlette and uvicorn (only required for the HTTP transports — included in requirements.txt)

Installation

git clone <repo-url> /opt/zabbix-mcp
cd /opt/zabbix-mcp

# Option A — system Python (no venv)
pip install -r requirements.txt

# Option B — virtual environment (recommended)
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# Option C — Docker
docker build -t zabbix-mcp:latest .

Configuration

Copy .env.example to .env and fill in your values:

cp .env.example .env

Zabbix connection

Variable	Required	Description
`ZABBIX_URL`	yes	Full base URL, e.g. `https://zabbix.example.com`
`ZABBIX_TOKEN`	one of†	API token — create at Administration → Users → API tokens
`ZABBIX_USER`	one of†	Username for user/password auth
`ZABBIX_PASSWORD`	one of†	Password for user/password auth
`ZABBIX_VERIFY_SSL`	no	`true` (default) / `false` to skip TLS verification

† Token auth takes priority when both are present.

LLM provider (agent only)

Variable	Description
`ZABBIX_MCP_PROVIDER`	Active provider: `nvidia` (default), `openrouter`, `groq`
`ZABBIX_MCP_NVIDIA_API_KEY`	NVIDIA Build API key — build.nvidia.com
`ZABBIX_MCP_NVIDIA_MODEL`	Model override (default: `meta/llama-3.3-70b-instruct`)
`ZABBIX_MCP_OPENROUTER_API_KEY`	OpenRouter API key
`ZABBIX_MCP_OPENROUTER_MODEL`	Model override (default: `openrouter/auto`)
`ZABBIX_MCP_GROQ_API_KEY`	Groq API key
`ZABBIX_MCP_GROQ_MODEL`	Model override (default: `llama-3.3-70b-versatile`)

Running the server

stdio transport (default)

The server reads JSON-RPC 2.0 from stdin and writes to stdout. It is normally spawned by an MCP client (Claude Desktop, an agent, etc.).

python server.py
# or explicitly
python server.py --transport stdio
# or with venv
.venv/bin/python server.py
# or with Docker
docker run --rm -i --env-file .env zabbix-mcp:latest

Streamable HTTP transport (recommended for remote clients)

The server starts a Starlette/uvicorn HTTP server exposing a single MCP endpoint at /mcp: clients POST their requests there, open the server-to-client stream with GET, and end the session with DELETE.

# local
python server.py --transport http --host 0.0.0.0 --port 8000
# or via env vars
MCP_TRANSPORT=http MCP_PORT=8000 python server.py
# or with Docker
docker run --rm --env-file .env -e MCP_TRANSPORT=http -p 8000:8000 zabbix-mcp:latest
# or via Make
make run-http
make run-http PORT=9000

Client URL: http://<host>:<port>/mcp Health check: GET http://<host>:<port>/health

--transport streamable-http is accepted as a synonym of --transport http.

HTTP+SSE transport (deprecated)

The HTTP+SSE transport of MCP 2024-11-05 (GET /sse + POST /messages/) was superseded by Streamable HTTP in the 2025-03-26 revision of the specification. It is kept for backwards compatibility: whenever the HTTP server runs, the legacy endpoints are served alongside /mcp, so old and new clients can both talk to the same process.

# Starts exactly the same server as --transport http, printing a deprecation
# warning. Legacy clients keep using http://<host>:<port>/sse
python server.py --transport sse --port 8000

To serve only the modern endpoint, disable the legacy routes:

python server.py --transport http --no-legacy-sse
# or: MCP_LEGACY_SSE=false

Transport CLI flags (HTTP mode):

Flag	Default	Description
`--path PATH`	`/mcp`	Path of the Streamable HTTP endpoint
`--json-response`	off	Reply to POSTs with plain JSON instead of an SSE stream
`--stateless`	off	Fresh transport per request, no session tracking (good behind a load balancer)
`--no-legacy-sse`	off	Do not expose `/sse` and `/messages/`

Transport environment variables:

Variable	Default	Description
`MCP_TRANSPORT`	`stdio`	`stdio`, `http` / `streamable-http`, or `sse` (deprecated alias of `http`)
`MCP_HOST`	`0.0.0.0`	Bind address (HTTP mode only)
`MCP_PORT`	`8000`	TCP port (HTTP mode only)
`MCP_PATH`	`/mcp`	Streamable HTTP endpoint path
`MCP_JSON_RESPONSE`	`false`	JSON instead of SSE replies
`MCP_STATELESS`	`false`	Stateless Streamable HTTP
`MCP_LEGACY_SSE`	`true`	Also serve the deprecated `/sse` + `/messages/`

See DOCKER.md for full Docker usage and Claude Desktop integration.

Running the agent

python agent.py

Zabbix AI Agent  |  provider=nvidia  model=meta/llama-3.3-70b-instruct
Type your request, or 'exit' / Ctrl-C to quit.

You: Show me all active problems with severity High or above
  [tool] get_problems({"min_severity": 4})
...

Available tools

Hosts

Tool	Description
`list_hosts`	List hosts with status, availability and primary IP. Optional filters: `groupid`, `status`, `limit`.
`get_host`	Full detail for one host: interfaces, groups, templates, macros. Accepts `hostid` or `hostname`.
`search_hosts`	Substring search across host name and visible name. Required: `query`.

Triggers / Problems

Tool	Description
`get_problems`	Active (unresolved) problems, sorted by severity desc. Optional filters: `min_severity`, `hostid`, `groupid`, `limit`.
`get_triggers`	Triggers for a host, sorted by severity. Required: `hostid`. Optional: `only_problems`, `min_severity`.

Items / History

Tool	Description
`get_items`	Monitoring items for a host with last collected value and unit. Required: `hostid`. Optional: `search`, `limit`.
`get_history`	Recent data points for one item. Required: `itemid`. Optional: `hours` (default 1), `limit` (default 100).

Maintenance

Tool	Description
`get_maintenances`	List all maintenance windows with scope and active period.
`create_maintenance`	Create a one-time maintenance window. Required: `name`, `hostids`, `start_time` (ISO 8601), `duration_minutes`.
`delete_maintenance`	Delete a maintenance window by `maintenanceid`.

Host Groups

Tool	Description
`get_host_groups`	List all host groups with IDs and names. Optional: `search`. Use `groupid` results to filter `list_hosts`, `get_problems`, etc.

Events

Tool	Description
`get_events`	Historical events (problems + recoveries) for a time window, including hostname. Optional: `hours` (default 24), `hostid`, `groupid`, `min_severity`, `limit`.
`acknowledge_problem`	Acknowledge one or more events by `eventids`. Optional: `message`, `close`.
`get_top_hosts_by_problems`	Ranked list of hosts by problem count in a time window. Optional: `hours` (default 168), `min_severity`, `top_n` (default 10).

Graphs

Tool	Description
`get_graphs`	Graphs defined for a host. Required: `hostid`. Optional: `search`.
`get_graph_items`	Items (metrics) that make up a graph. Required: `graphid`.

Templates

Tool	Description
`get_templates`	List templates. Optional: `search`, `hostid` (filter by linked host).

Inventory

Tool	Description
`get_host_inventory`	Full inventory record for a host (OS, hardware, location, serial numbers). Required: `hostid`.

Actions

Tool	Description
`get_actions`	List alerting actions with status and event source. Optional: `search`.

Analytics

Tool	Description
`get_flapping_triggers`	Triggers with most state changes (PROBLEM↔OK) in a window — identifies noisy thresholds. Optional: `hours`, `hostid`, `groupid`, `min_flaps` (default 2), `top_n`.
`get_availability_report`	Uptime % per host computed from PROBLEM events and their recoveries. Optional: `hours` (default 720 = 30d), `hostids`, `groupid`, `min_severity` (default 3).
`get_trends`	Hourly aggregated trend data (min/avg/max) for an item over days/months. Required: `itemid`. Optional: `hours` (default 720), `limit`.
`get_problem_duration_stats`	Per-trigger stats: count, total/avg/max downtime. Optional: `hours`, `hostid`, `groupid`, `min_severity`, `top_n`.

Operational

Tool	Description
`enable_host`	Enable monitoring for a host. Required: `hostid`.
`disable_host`	Disable monitoring for a host. Required: `hostid`.
`update_maintenance`	Update an existing maintenance window (name, time, duration, collect_data). Required: `maintenanceid`.
`add_host_to_group`	Add a host to a host group. Required: `hostid`, `groupid`.
`remove_host_from_group`	Remove a host from a host group. Required: `hostid`, `groupid`.

Users / Groups

Tool	Description
`get_users`	List users with username, display name and group membership. Optional: `search`.
`get_user_groups`	List user groups with GUI access level and status. Optional: `search`.

Severity codes

Code	Label
0	Not classified
1	Information
2	Warning
3	Average
4	High
5	Disaster

Project structure

zabbix-mcp/
├── server.py          — MCP server (30 tools, stdio + Streamable HTTP + legacy SSE)
├── util.py            — Zabbix code → human-readable string converters
├── agent.py           — Interactive CLI agent
├── llm.py             — LLM provider registry and agentic loop
├── requirements.txt   — Python dependencies
├── Dockerfile         — Container image for the MCP server
├── DOCKER.md          — Docker usage and Claude Desktop integration guide
├── .env.example       — Environment variable template
└── .gitignore

How the agentic loop works

User question
      │
      ▼
  LLM (Reason)  ──── tool_calls? ──► MCP server (Act)
      ▲                                     │
      │                                     │ result
      └─────────── tool message ◄───────────┘
                  (Observe)
      │
  finish_reason = "stop"
      │
      ▼
  Final answer

The loop runs up to 20 iterations by default and handles unknown tools, JSON decode errors, and asyncio timeouts gracefully.

Integrating with Claude Desktop

stdio mode (subprocess — no port needed)

{
  "mcpServers": {
    "zabbix": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "--env-file", "/path/to/.env", "zabbix-mcp:latest"]
    }
  }
}

Without Docker:

{
  "mcpServers": {
    "zabbix": {
      "command": "/opt/zabbix-mcp/.venv/bin/python",
      "args": ["/opt/zabbix-mcp/server.py"]
    }
  }
}

Streamable HTTP mode (remote/persistent container)

First start the container:

docker run -d --name zabbix-mcp \
  --env-file /path/to/.env \
  -e MCP_TRANSPORT=http \
  -p 8000:8000 \
  zabbix-mcp:latest

Then add to claude_desktop_config.json:

{
  "mcpServers": {
    "zabbix": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

Or register it with the Claude Code CLI:

claude mcp add --transport http zabbix http://localhost:8000/mcp

Legacy SSE mode (deprecated clients only)

The same container also answers on the old endpoint, so a client that cannot speak Streamable HTTP yet keeps working unchanged:

{
  "mcpServers": {
    "zabbix": {
      "url": "http://localhost:8000/sse"
    }
  }
}

Prefer /mcp for anything new — /sse will be removed once clients have caught up.

License

MIT

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained

ResponsivenessNo issues

Resources

GitHub Repository

Need Help?

Related Servers

Related MCP Connectors

Backtest360 A
com.backtest360
MCP server exposing the Backtest360 engine API as tools for AI agents.
Finance Research & Data
202
mcp-serverOAuth
com.niftypm
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
Project Management Communication File Systems
2
ZapierOAuth
com.zapier
Hosted MCP server connecting AI assistants to 9,000+ apps and 40,000+ actions via Zapier.
App Automation Workplace & Productivity
2297
nexusOAuth
com.civic
Zero-setup MCP gateway securely connecting AI to your tools with authentication and workflows
Agent Orchestration Developer Tools App Automation
2

Related MCP Servers

zabbix-mcp-server
Monitoring Agent Orchestration
mpeirone
A
license
A
quality
C
maintenance
🔌 Complete MCP server for Zabbix integration - Connect AI assistants to Zabbix monitoring with 40+ tools for hosts, items, triggers, templates, problems, and more. Features read-only mode and comprehensive API coverage.
Updated 5 months ago (2026-05-10 12:29 UTC)
3
255
GPL 3.0
mcp-zabbix
Monitoring Observability API Testing
daedalus
A
license
C
quality
C
maintenance
Exposes the complete Zabbix API functionality through the Model Context Protocol, mapping API methods to tools for managing hosts, triggers, and monitoring data. It enables seamless integration and control of Zabbix monitoring environments via natural language interfaces.
Updated 5 months ago (2026-04-21 19:29 UTC)
100
MIT
Zabbix MCP Server
Monitoring Observability
initMAX
A
license
Not graded
quality
A
maintenance
Exposes the complete Zabbix API to MCP-compatible AI assistants, enabling natural language management of hosts, problems, and templates across multiple instances. It provides 220 tools for comprehensive monitoring and configuration with support for read-only modes and secure authentication.
Updated 2 months ago (2026-08-07 20:10 UTC)
206
AGPL 3.0
zabbix-investigation-mcp
Monitoring Observability
Edrient17
F
license
Not graded
quality
B
maintenance
Read-only MCP server that provides Zabbix monitoring data to AI agents, enabling investigation of hosts, metrics, incidents, and triggers with deterministic aggregation.
Updated a month ago (2026-08-25 08:56 UTC)
-

zabbix-mcp

Features

Requirements

Installation

Configuration

Zabbix connection

LLM provider (agent only)

Running the server

stdio transport (default)

Streamable HTTP transport (recommended for remote clients)

HTTP+SSE transport (deprecated)

Running the agent

Available tools

Hosts

Triggers / Problems

Items / History

Maintenance

Host Groups

Events

Graphs

Templates

Inventory

Actions

Analytics

Operational

Users / Groups

Severity codes

Project structure

How the agentic loop works

Integrating with Claude Desktop

stdio mode (subprocess — no port needed)

Streamable HTTP mode (remote/persistent container)

Legacy SSE mode (deprecated clients only)

License

Maintenance

Resources

Related MCP Connectors

Related MCP Servers

zabbix-mcp-server

mcp-zabbix

Zabbix MCP Server

zabbix-investigation-mcp