Matomo-MCP
Matomo MCP Server to Work fast with Analytics
README
<div align="center">
<img src="assets/logo.svg" width="120" alt="matomo-mcp logo"/>
matomo-mcp
Talk to your Matomo Analytics. From Claude, Cursor, VS Code, or any MCP client.
14 curated, read-only analytics tools + a full-API escape hatch. Single binary, instant startup, context-friendly.
Quickstart · Clients · Tools · Configuration · FAQ
</div>
You ▸ How was traffic yesterday, and where did it come from?
Claude ▸ Yesterday you had 14,472 visits (11,416 unique visitors, 66% bounce rate).
Top acquisition channels:
1. Organic search — 6,120 visits (Google 92%)
2. Direct — 4,890 visits
3. AI assistants — 1,204 visits (↑ 31% vs. last week)
Want me to break down which landing pages converted best?
Every question your Matomo dashboard can answer, your AI assistant can now answer too — including follow-ups, comparisons, and "why?".
✨ Why matomo-mcp?
| 🎯 Curated, not generated | 14 hand-crafted tools modeled on real analytics questions — not 70+ auto-generated API mirrors that flood the model's context and degrade tool selection. |
| ⚡ Instant startup | No introspection round-trips. One static binary, no Node, no Python, no runtime. Starts in milliseconds. |
| 🔒 Safe by default | Read-only reporting tools. Token sent via POST only (never in URLs/logs), redacted from every error. TLS verification on by default. |
| 🧠 Context-friendly | Row limits on every report and a hard response budget with actionable guidance — one tool call can never blow up the context window. |
| 📡 Real-time included | Live visitor counters and a visit log (matomo_realtime) — see what's happening right now. |
| 🧰 Never a cage | matomo_api reaches any Reporting API method (funnels, heatmaps, custom dimensions, …) when the curated tools don't cover it. |
| 🔁 Resilient | Automatic retries with backoff on 429/5xx/network hiccups. Helpful, hint-annotated error messages the model can act on. |
🚀 Quickstart
1. Install
Prebuilt binary (Linux, macOS, Windows) — grab it from Releases, or:
# Cargo
cargo install matomo-mcp
# From source
cargo install --git https://github.com/Liohtml/matomo-mcp
# Docker
docker pull ghcr.io/liohtml/matomo-mcp
2. Get a Matomo API token
Matomo → Settings (⚙) → Personal → Security → Auth tokens → Create new token. View-only permissions are all it needs.
3. Verify the connection
matomo-mcp --url https://your-matomo.example.com --token YOUR_TOKEN --check
✓ Connected — Matomo version 5.2.1
✓ Token grants access to 3 site(s):
#1 My Shop (https://shop.example.com)
#2 Blog (https://blog.example.com)
#3 Docs (https://docs.example.com)
4. Connect your client ⬇
🔌 Connect your client
<details> <summary><b>Claude Code</b></summary>
claude mcp add matomo \
--env MATOMO_URL=https://your-matomo.example.com \
--env MATOMO_TOKEN=YOUR_TOKEN \
--env MATOMO_DEFAULT_SITE_ID=1 \
-- matomo-mcp
</details>
<details> <summary><b>Claude Desktop</b></summary>
Add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):
{
"mcpServers": {
"matomo": {
"command": "matomo-mcp",
"env": {
"MATOMO_URL": "https://your-matomo.example.com",
"MATOMO_TOKEN": "YOUR_TOKEN",
"MATOMO_DEFAULT_SITE_ID": "1"
}
}
}
}
</details>
<details> <summary><b>Cursor</b></summary>
.cursor/mcp.json (project) or ~/.cursor/mcp.json (global):
{
"mcpServers": {
"matomo": {
"command": "matomo-mcp",
"env": {
"MATOMO_URL": "https://your-matomo.example.com",
"MATOMO_TOKEN": "YOUR_TOKEN",
"MATOMO_DEFAULT_SITE_ID": "1"
}
}
}
}
</details>
<details> <summary><b>VS Code (GitHub Copilot)</b></summary>
.vscode/mcp.json:
{
"servers": {
"matomo": {
"type": "stdio",
"command": "matomo-mcp",
"env": {
"MATOMO_URL": "https://your-matomo.example.com",
"MATOMO_TOKEN": "${input:matomo-token}",
"MATOMO_DEFAULT_SITE_ID": "1"
}
}
},
"inputs": [
{
"id": "matomo-token",
"type": "promptString",
"description": "Matomo API token",
"password": true
}
]
}
</details>
<details> <summary><b>Windsurf / Zed / other MCP clients</b></summary>
Any client that speaks MCP over stdio works with the generic shape:
{
"command": "matomo-mcp",
"args": [],
"env": {
"MATOMO_URL": "https://your-matomo.example.com",
"MATOMO_TOKEN": "YOUR_TOKEN",
"MATOMO_DEFAULT_SITE_ID": "1"
}
}
</details>
<details> <summary><b>Docker (any client)</b></summary>
{
"mcpServers": {
"matomo": {
"command": "docker",
"args": [
"run", "-i", "--rm",
"-e", "MATOMO_URL", "-e", "MATOMO_TOKEN", "-e", "MATOMO_DEFAULT_SITE_ID",
"ghcr.io/liohtml/matomo-mcp"
],
"env": {
"MATOMO_URL": "https://your-matomo.example.com",
"MATOMO_TOKEN": "YOUR_TOKEN",
"MATOMO_DEFAULT_SITE_ID": "1"
}
}
}
}
</details>
[!TIP] Set
MATOMO_DEFAULT_SITE_IDand the model never has to ask which site you mean. No token at hand? Try it against the public demo:--url https://demo.matomo.cloud --default-site-id 1(no token needed).
🧭 Tools
| Tool | Answers questions like |
|---|---|
matomo_list_sites |
"Which sites do we track?" |
matomo_visits_summary |
"How much traffic did we get last week?" |
matomo_pages |
"What are our top pages? Where do people exit?" |
matomo_referrers |
"Where do visitors come from? Which campaigns work? What do AI assistants send us?" |
matomo_events |
"How often was the configurator opened?" |
matomo_goals |
"What's our conversion rate per goal?" |
matomo_ecommerce |
"Revenue this month? Best-selling products?" |
matomo_geo |
"Which countries/cities do visitors come from?" |
matomo_devices |
"Mobile vs. desktop? Which browsers?" |
matomo_visit_times |
"When during the day/week do people visit?" |
matomo_site_search |
"What do people search for on our site — and find nothing?" |
matomo_realtime |
"Who's on the site right now?" |
matomo_page_performance |
"Which pages load slowly?" |
matomo_api |
Everything else — funnels, heatmaps, custom dimensions, any Module.action of the Reporting API |
All tools accept site_id, period (day/week/month/year/range), date
(today, yesterday, 2026-07-01, last30, or start,end ranges), an optional
segment (e.g. deviceType==mobile;country==DE), and a row limit.
Prompts to try
- "Compare this week's traffic with last week — what changed and why?"
- "Top 10 landing pages by conversions this month, with bounce rates."
- "Are we getting traffic from ChatGPT or Perplexity? Trend over 3 months."
- "Which internal searches return no results? Suggest content we should create."
- "Anything unusual in the visitor log right now?"
⚙️ Configuration
| Flag | Env | Default | Description |
|---|---|---|---|
--url |
MATOMO_URL |
— (required) | Matomo instance URL (sub-directory installs like https://example.com/matomo/ work) |
--token |
MATOMO_TOKEN |
— | API token (token_auth), view access is enough |
--default-site-id |
MATOMO_DEFAULT_SITE_ID |
— | Site used when the model doesn't specify one |
--header |
MATOMO_EXTRA_HEADERS |
— | Extra HTTP headers (Name:Value, repeatable / comma-separated) — for auth proxies, Zero-Trust, multi-tenant setups |
--timeout-secs |
MATOMO_TIMEOUT_SECS |
30 |
Per-request timeout |
--max-response-chars |
MATOMO_MAX_RESPONSE_CHARS |
50000 |
Response budget before truncation |
--insecure |
MATOMO_INSECURE |
false |
Accept self-signed TLS certificates (explicit opt-in) |
--check |
— | — | Verify URL + token + site access, then exit |
🆚 How is this different from FGRibreau/mcp-matomo?
mcp-matomo (which inspired this project — thanks! 🙏) introspects your Matomo instance at startup and generates one MCP tool per API method. matomo-mcp takes the opposite approach:
| matomo-mcp | mcp-matomo | |
|---|---|---|
| Tool set | 14 curated tools + escape hatch | ~70+ generated tools |
| Model context cost | Small, stable | Large, instance-dependent |
| Parameter types | Exact, hand-written enums/defaults | Inferred from parameter names |
| Startup | Instant (no network I/O) | Introspection round-trips (or cached spec file) |
| TLS verification | On by default | Disabled for introspection |
| Sub-directory installs | ✅ | Path is overwritten |
| Response size guard | Row limits + hard budget | — |
| Retries on transient errors | ✅ | — |
| Real-time (Live) tools | ✅ | — (not part of report metadata) |
If you want every API method as its own tool, use mcp-matomo. If you want the model to reliably pick the right tool and never flood its context, use matomo-mcp.
🩺 Troubleshooting
<details> <summary><b>"site_id is required"</b></summary>
Either pass --default-site-id 1 (recommended) or let the model call matomo_list_sites first.
</details>
<details> <summary><b>401 / "cannot be authenticated"</b></summary>
Run matomo-mcp --url ... --token ... --check. If it fails: regenerate the token (Settings → Personal → Security), make sure it has at least view access to the site.
</details>
<details> <summary><b>404 or HTML instead of JSON</b></summary>
MATOMO_URL must point at the Matomo root — the folder containing index.php. For https://example.com/matomo/index.php, use https://example.com/matomo/.
</details>
<details> <summary><b>Behind Cloudflare Access / OAuth2 proxy / Zero Trust?</b></summary>
Inject the bypass headers: --header "CF-Access-Client-Id:..." --header "CF-Access-Client-Secret:..." (or via MATOMO_EXTRA_HEADERS).
</details>
<details> <summary><b>Responses feel truncated</b></summary>
That's the context guard doing its job. Ask for fewer rows, a shorter date range, or raise --max-response-chars.
</details>
🗺️ Roadmap
- [ ] Streamable HTTP transport (host it once, connect many clients)
- [ ]
matomo_annotations— read & correlate deploy markers with traffic - [ ] Multi-instance support (one server, several Matomo installations)
- [ ] Homebrew tap & winget manifest
- [ ] MCP registry listing
Want one of these sooner? Open an issue — or a PR, see CONTRIBUTING.md.
🛠️ Development
cargo test # 33 tests, fully offline (wiremock)
cargo clippy --all-targets -- -D warnings
cargo run -- --url https://demo.matomo.cloud --default-site-id 1 --check
Architecture and design decisions: docs/ARCHITECTURE.md.
📄 License & Credits
MIT. Not affiliated with or endorsed by Matomo — Matomo is a registered trademark of InnoCraft Ltd.
Built with rmcp, the official Rust MCP SDK. Inspired by FGRibreau/mcp-matomo.
- MCP Registry name:
mcp-name: io.github.Liohtml/matomo-mcp
<div align="center">
If matomo-mcp saves you a dashboard visit, a ⭐ helps others find it.
</div>
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。