MCP (AI assistants)

MCP (AI assistants)

MetriXs exposes a read-only MCP (Model Context Protocol) endpoint so you can query your analytics from inside an AI assistant: “how many visitors did my site have this week, and where did they come from?”, answered with real data instead of a guess.

Endpoint: https://mcp.metrixs.eu/api/mcp

MetriXs never calls an AI provider itself. You connect your own assistant with your own credentials; only aggregated analytics numbers flow to the chat you chose. Nothing leaves your MetriXs instance except the answers to the questions you ask in it.

Agencies: clients cannot connect the MCP server

Client accounts on Agency plans cannot use the MCP server: the consent flow refuses customer accounts outright, and clients cannot create API keys in the dashboard either. If a client wants AI access to their analytics, the agency creates a site API key (Settings → Sites → API keys) and shares it with them — the key stays pinned to that one site and read-only.

Two ways to connect

1. API key (Claude Code, Codex CLI, Gemini CLI, VS Code / GitHub Copilot, Cursor, JetBrains)

1. In the dashboard, open Settings → Sites and create an API key for the site you want to share. The key is shown once.
2. Add the server to your client with the URL above and the header Authorization: Bearer <key>:

Client	Where
Claude Code	claude mcp add metrixs --transport http https://mcp.metrixs.eu/api/mcp --header "Authorization: Bearer <key>"
VS Code / GitHub Copilot	.vscode/mcp.json → servers.metrixs with type: "http", the URL and the header
Codex CLI	~/.codex/config.toml → [mcp_servers.metrixs] with url + bearer_token_env_var
Gemini CLI	~/.gemini/settings.json → mcpServers.metrixs with httpUrl + headers
Cursor	MCP settings → add a custom (HTTP) server with the URL and header
JetBrains AI Assistant	Settings → MCP, add a remote server with the URL and header

2. OAuth (Claude Desktop, ChatGPT connectors, Cursor, JetBrains)

1. In your assistant, add a custom remote MCP server / connector with the URL above. Claude Desktop and ChatGPT only support this flow; they do not accept plain API keys.
2. The assistant opens a MetriXs page in your browser. Log in, pick one site, and allow access.
3. Done. The assistant can now read that site’s analytics.

Every OAuth connection appears as an API key named MCP: <assistant> under Settings → Sites → API keys. Deleting that row revokes the connection instantly, including live sessions.

Available tools (all read-only)

list_sites, get_current_visitors, get_overview, get_timeseries, get_breakdown (pages, sources, countries, devices, UTM, …), get_events, get_event_props, get_revenue, get_funnel, get_top_products (commerce sites), get_ai_traffic, get_blocked_bots, get_security_audit (latest security audit, Team, Pro, and Agency plans).

Privacy

• Read-only: no tool can change settings, sites or data.
• One site per connection: a key is pinned to a single site; assistants never see past it.
• Aggregates only: no visitor identifiers exist to expose; visitor IDs are daily-rotating hashes and are never included in any tool response.
• EU-hosted: MetriXs makes no AI calls itself. Your assistant talks to MetriXs; MetriXs never talks to an AI provider.