MCP analytics server

Privatus Analytics runs a hosted Model Context Protocol server. The API operations you're allowed to use are MCP tools, so an agent can read your stats, create goals, add notes or check uptime with the same permissions as your API token. The tools are listed under All tools.

Set it up in your client: Claude, ChatGPT, Gemini CLI, Cursor and VS Code.

Authentication#

API token. Create a token in Workspace settings → API tokens and send it as a Bearer token. Give the token only the permissions the agent needs. For read-only analysis, analytics.read is enough. See Authentication.

OAuth. Clients that support MCP authorization can connect with just the URL. The server publishes its metadata at /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server, supports dynamic client registration (/oauth/register), client ID metadata documents (the way Claude, ChatGPT and Smithery name themselves) and the authorization code flow with PKCE (S256 only). You sign in, choose the workspace and the access, and approve:

The client receives an API token named after the client, with "(MCP)" at the end. It appears in your token list, where you can revoke it. The access token lasts an hour, and the client renews it with a refresh token (grant_type=refresh_token), which is replaced at every renewal. Each renewal reads your role again, so a token narrows when your role does and stops when you leave the workspace. A connection nobody used for 90 days ends, and the client asks you to approve again. Revoking either token at /oauth/revoke ends the connection.

Clients that send a resource parameter (RFC 8707) must name this server's /mcp address, or the request is refused with invalid_target. Every response sent back to the client's redirect address carries iss (RFC 9207).

A request without a valid token gets 401 with a WWW-Authenticate header that points to the metadata. A request from a web page of another site (an Origin header that isn't ours or a local address) gets 403.

Tool names#

Tools are named <resource>_<verb>:

Each tool's input schema is the operation's inputs plus its path parameters (site_id, workspace_id…). Every tool has a title and all four annotations, so clients can decide which tools to ask about first:

Some tools are for a person and only return a link.

The server only lists tools your token's permissions allow. A token limited to some sites isn't offered the workspace-wide tools. Tool calls go through the same routes, permission checks and validation as the API.

Actions for a person#

An AI assistant can pick the wrong tool or the wrong id, so the actions that are hard to undo are left to you. Each has a tool named for the page it returns (sites_delete_page, "Get the app page to delete a site"). Calling it changes nothing: it responds with the code human_only and a link to the page of the app where you do it yourself, which the assistant passes on to you. They are:

• Deleting collected data: a site, a date range of its data, a workspace, a custom event, an import, an export, a survey with its responses, and disconnecting Search Console, Bing, a payment provider or Cloudflare (which deletes what they imported).
• Billing: checkout, plan changes, canceling and resuming, billing settings, the badge bonus, discount applications and the Paddle links.
• Access and security: API tokens, invitations, members, custom roles, per-site access, ownership and site transfers, SSO, workspace security and crawler log keys.
• Warehouse destinations: adding, changing, deleting and syncing.
• Testimonials: everything that deletes a record, moves it to the trash or empties the trash. An assistant cannot delete anything or move anything to the trash. It hands you a link to the page. Bulk actions on many testimonials at once are not a tool (each action has its own, and the bulk route stays in the JSON API), and an undo is refused when taking the run back would delete or trash something (the undo of a restore, a translate run or a make images run). See Trash.

The JSON API still runs every one of them, because a call there is written on purpose. In the API reference they are marked HTTP only, and in All tools the summary says so.

An email when data is sent out#

Creating an export, revoking a share link, creating or revoking a wallboard link, and adding, changing, deleting, testing or redelivering a webhook all run over MCP. Each time an assistant does one of them, the member whose token it used gets an email that names the action and the token, with a link to the API tokens page. The same call over the API sends no email.

Nothing that takes a secret#

No tool takes a password, an API key or a token, because what you type in a chat stays in its transcript. The actions that take one are in the app only, never over MCP or the API: creating or changing a share link (it can have a password), adding or changing a notification channel, connecting Cloudflare, Search Console or Bing, a subscription provider, a purchase connection (Stripe or Paddle) and a move from another testimonials tool. Reading a share link or a site transfer by its token is in the JSON API only.

Each request to /mcp counts as one request toward the workspace's API rate limit. Over the limit you get 429 with the error code rate_limited and a Retry-After header.

Start here#

Agents usually begin with:

1. workspaces_list → your workspace id (ws_…).
2. sites_list with workspace_id → site ids (pa_…).
3. stats_aggregate, stats_timeseries, stats_breakdown or stats_overview with site_id, period and filters (filter syntax).
4. docs_search and docs_get to read these docs.

See Examples.

All tools#

Generated from the running code. The API reference has each tool's full input and output schema.

API tokens#

Account#

Ad creatives#

Ad placements#

Affiliate program#

Alerts#

Analytics#

Assistant#

Audit log#

Billing#

Bot detection#

Campaigns#

Channel rules#

Crawlers#

Dashboards#

Documentation#

Domain Rating#

Email reports#

Events#

Experiments#

Exports#

Forms#

Funnels#

Goals#

Help and feedback#

Imports#

Live#

Members#

Notes#

Notification channels#

Performance#

Plans#

Private feedback#

Removal requests#

Rewards#

Roles#

SSO#

Search#

Segments#

Sharing#

Site audit#

Sites#

Sponsor reports#

Status pages#

Subscriptions#

Surveys#

Testimonial analytics#

Testimonial brand kit#

Testimonial collections#

Testimonial companies#

Testimonial custom domain requests#

Testimonial custom properties#

Testimonial duplicates#

Testimonial images#

Testimonial import runs#

Testimonial import sources#

Testimonial languages#

Testimonial media#

Testimonial mentions#

Testimonial migrations#

Testimonial people#

Testimonial purchase connections#

Testimonial request recipients#

Testimonial requests#

Testimonial saved views#

Testimonial spreadsheet imports#

Testimonial tags#

Testimonial translations#

Testimonial trash#

Testimonial undo list#

Testimonials#

Tracking#

Traffic rules#

Uptime#

Walls of Love#

Warehouse sync#

Webhooks#

Website#

Widgets#

Workspaces#