Analytics for AI Assistants with MCP
HitKeep MCP gives approved assistants and internal tools read-only access to aggregate analytics without sharing dashboard cookies, raw hit exports, or admin permissions. Use it when you want analytics for AI assistants from Claude, Cursor, or another MCP client, but still need the dashboard, API, reports, and exports governed by HitKeep permissions.
A governed analytics MCP server for read-only aggregate analytics from approved AI assistants and internal reporting tools.
Best fit
Section titled “Best fit”Use this path when you want:
- an AI assistant to answer analytics questions from approved HitKeep data
- Claude, Cursor, or another MCP client to ask traffic, event, ecommerce, Web Vitals, AI visibility, and Search Console questions
- internal reporting tools that can query aggregate site metrics without dashboard access
- scoped access by site through API client tokens
- ecommerce, event, Web Vitals, AI visibility, Search Console, and overview analytics through one governed interface
- official HitKeep docs lookup beside analytics answers
MCP is not the right surface for raw data export, site administration, billing, alert creation, funnel mutation, or other write workflows. Use open exports and takeout for portable files and the REST API reference for normal API automation.
How access is governed
Section titled “How access is governed”MCP is disabled by default. When enabled, HitKeep mounts a Streamable HTTP endpoint on the existing HTTP server, usually at /mcp.
hitkeep \ -mcp-enabled \ -mcp-path /mcpMCP accepts existing API client bearer tokens only. It does not accept dashboard cookies. The token must be scoped to a visible site and must pass the same site.view permission checks used by the dashboard and REST API.
What assistants can ask
Section titled “What assistants can ask”The v1 MCP surface is read-only and aggregate-only.
| Question | MCP tool surface |
|---|---|
| Which sites can this token see? | hitkeep_list_sites |
| How did a site perform over a date range? | hitkeep_get_site_overview |
| Which events fired, and what properties did they carry? | hitkeep_get_event_names, hitkeep_get_event_breakdown |
| Which products, orders, and sources drove ecommerce activity? | hitkeep_get_ecommerce |
| Which pages have poor p75 Web Vitals? | hitkeep_get_web_vitals |
| Which AI systems fetched or referred traffic? | hitkeep_get_ai_visibility |
| Which saved recommendations should this site review? | hitkeep_get_opportunities |
| Is Search Console mapped and synced for this site? | hitkeep_get_search_console_status |
| Which imported queries, pages, countries, or devices drove organic search clicks? | hitkeep_get_search_console |
| What does the official HitKeep documentation say? | hitkeep_search_docs, hitkeep_get_doc, hitkeep_get_api_reference |
Date ranges default to the last 30 days and are capped by -mcp-max-range-days. Search Console MCP tools read imported aggregate rows only. They do not call Google live or trigger a sync. See the configuration reference for MCP flags and defaults.
Cloud or self-hosted
Section titled “Cloud or self-hosted”Use HitKeep Cloud when the main blocker is operating a public endpoint, TLS, updates, SMTP, or backups. Use self-hosted HitKeep when you need direct control over the runtime, token lifecycle, reverse proxy, and network policy.
Either way, keep MCP tokens narrow. Create team-owned API clients for shared assistants, scope them to only the sites they need, and revoke them when the tool no longer needs access.