Skip to main content
Alpic is conservative by default: analytics work out of the box on metadata alone, and anything that could contain end-user content is opt-in.

​
What is collected by default

For every MCP request, Alpic records metadata only:
  • The method and tool name called
  • The MCP client name (ChatGPT, Claude, Cursor, …)
  • Timestamp, response time, and estimated output tokens
  • The error type and code, if the request failed
  • A user identity when one can be resolved, stored as a hash for keys and tokens (see How users are counted)
What is never collected, on any plan:
  • IP addresses
  • Raw API keys or tokens, only hashes are stored
  • Request or response contents, unless you explicitly enable payload collection below

​
Data collection settings

Privacy settings
Data collection is configured in Project Settings → Privacy, separately for each environment. This lets you, for example, collect payloads on staging while keeping production metadata-only. All of these settings require a paid plan. On the Free plan they appear locked, and nothing beyond metadata is ever stored, including error messages.
Payloads and error messages may contain end-user PII. You are responsible for what your server puts in them. Review your obligations in ourPrivacy Policybefore enabling collection.
A few practical notes:
  • The coarse location is only what ChatGPT itself sends with the call: country, city, and coordinates rounded to two decimals (about 1.1 km). It is never derived from the caller’s IP address, and calls from other MCP clients carry no location at all.
  • Payloads are collected only for tools/call, prompts/get, and resources/read requests, and only while both the corresponding toggle and session replay are enabled.
  • Payloads larger than 256 KB are skipped.
  • Collected data follows your plan’s analytics retention.