Matomo-MCP

Matomo-MCP

Matomo MCP Server to Work fast with Analytics

Category
Visit Server

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.

CI Crates.io License: MIT Rust MCP

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_ID and 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>

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured