mikrotik-mcp
Provides read-only (with limited address-list management) access to MikroTik RouterOS routers, enabling monitoring of system info, interfaces, DHCP leases, firewall rules, and more via natural language.
README
mikrotik-mcp
MCP server for MikroTik RouterOS. Talks to the router directly over the binary RouterOS API (port 8728) — a minimal hand-rolled implementation of the protocol (length encoding, sentence-based exchange, MD5-challenge login for older versions), no third-party RouterOS API library.
Intentionally read-only (with one exception — managing entries in a single address-list, see below). This is a deliberate architectural choice: the router is the most sensitive node in the network, and an LLM shouldn't be able to change firewall/NAT/routing directly.
Tools
| Tool | Description |
|---|---|
system_info |
Model, RouterOS version, uptime, CPU, RAM |
get_interfaces |
List of network interfaces with status |
get_dhcp_leases |
DHCP leases — who's connected to the network |
get_firewall_rules |
Firewall filter rules, filterable by chain |
get_nat_rules |
NAT rules (dstnat/srcnat) |
get_address_lists |
Contents of an address-list |
get_queues |
Simple Queues — bandwidth limits |
get_routes |
Routing table |
add_to_address_list / remove_from_address_list |
The only write operations — add/remove an IP from an address-list (e.g. for blocking) |
get_logs |
Recent log entries |
execute_command |
Arbitrary read-only RouterOS command — only from an explicit whitelist in the code (ALLOWED_READ_COMMANDS) |
get_wireguard |
WireGuard interfaces and peers (endpoint, handshake, rx/tx) |
get_dns_static |
Router's static DNS entries |
get_interface_traffic |
Instant per-interface throughput snapshot (/interface/monitor-traffic ... once) |
config_snapshot |
JSON snapshot of key config sections — for diffing before/after manual changes |
Setup
git clone <this-repo> mikrotik-mcp && cd mikrotik-mcp
python3 -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env # fill in MIKROTIK_HOST/USER/PASS, MCP_SECRET
uvicorn server:app --host 0.0.0.0 --port 8001
Systemd unit example: deploy/mikrotik-mcp.service.
Docker:
docker build -t mikrotik-mcp .
docker run -p 8001:8001 --env-file .env mikrotik-mcp
On the router: create a dedicated user in a read-only group (/user group add name=ai-mcp-group policy=read,api,!write,!policy,!test,!winbox,!password,!web,!reboot,!ftp,!sniff,!sensitive,!romon), don't grant write/policy. Even if the server itself is compromised, access stays limited at the RouterOS level.
Security model
- Auth is an
Authorization: Bearer $MCP_SECRETheader. EmptyMCP_SECRET= no check (local network/VPN only). /.well-known/oauth-authorization-server+/oauth/authorize+/oauth/tokenare a compatible stub for claude.ai custom connectors, which don't support a static API key — only full OAuth 2.1 or no auth at all. The actual protection is the Bearer token on/mcp, not this handshake. Via Claude Code CLI (claude mcp add --header ...) you don't need this stub at all.redirect_uriin/oauth/authorizeis checked against an allowlist (claude.ai,anthropic.com,console.anthropic.com,localhost).execute_commandruns strictly through a whitelist of root commands in the code — you cannot execute an arbitrary write command through this tool, even by trying.- Transport: the server does not terminate TLS itself — it listens on plain HTTP. If it's reachable beyond localhost/a trusted LAN (and especially if you're connecting it as a custom connector in claude.ai, where HTTPS is required), put TLS termination in front of it: Cloudflare Tunnel, Tailscale Funnel, nginx/Caddy + Let's Encrypt, etc. Without that, the Bearer token (
MCP_SECRET) in theAuthorizationheader goes out in plaintext.
Requirements
- MikroTik RouterOS 6.x/7.x with the API enabled (
/ip service enable api, port 8728 by default). - Python 3.11+.
License
MIT — see LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。