Understand product usage with AI

What it is

Instead of building dashboards for every question, you ask an AI agent that can query your product data directly, then check the answer against the underlying charts and replays.

How it works

1. Instrument the product with an analytics tool such as PostHog (events, funnels, retention).
2. Connect it to your agent: PostHog’s hosted MCP server works with MCP clients including Claude Code, Claude Desktop, Codex and Cursor, with a read-only option.
3. Ask specific questions: “Show the signup-to-paid funnel for the last 30 days by device.”
4. Verify with session replays and the raw chart before acting.
5. Act: run an experiment or roll out a fix behind a feature flag.

Free tier (checked 2026-10-16). PostHog’s free allowance resets monthly and includes 1 million analytics events, 5,000 session recordings and 1 million feature-flag requests, with experiments billed with feature flags, plus 500 PostHog AI credits. Without a card on file, usage stops at the free limits rather than being billed.

Benefits

• Faster answers without waiting on an analyst.
• Engineers can pull product context while they code.
• Experiments and flags close the loop from question to change.

Limits and risks

• An agent can misread a metric definition; check the query.
• Session replays capture user behavior; configure privacy masking.
• AI-powered tools cost extra. Connecting to PostHog’s MCP server and calling its tools is free, but some tools run LLMs internally (PostHog gives sentiment analysis and trace summarization as examples) and may be billed as PostHog AI spend. They only appear when AI data processing is on in your organization settings, so turning that off filters them out.