Connecting and calling tools costs nothing; normal PostHog usage billing still applies to the data behind your queries, and the tools that run an LLM internally bill as PostHog AI spend and only appear when AI data processing is enabled for the organisation. Because it writes as well as reads, the levers worth setting up front are read-only mode, tool filtering by name or category, and pinning the session to one organisation or project.
PostHog's free hosted MCP endpoint at https://mcp.posthog.com/mcp. It exposes PostHog's products as function-calling tools to any MCP-compatible client, and routes to the US or EU region based on the account you sign in with.
- Feature flags, experiments and surveys created, launched and updated from a prompt
- Error tracking: issue lists, a single issue with its events, merges, splits and assignment rules
- SQL run against your project, plus insights, dashboards and web analytics queries
- Logs queried, counted, pattern-mined and diffed between two time windows
- Session recordings, playlists, CDP destinations and data warehouse sources
- Stable tool names, so a tools query parameter can scope a session to a subset
- Npx @posthog/wizard mcp add installs it into PostHog Desktop, Cursor, Claude Code, Claude Desktop, Codex, VS Code and Zed
A PostHog account. OAuth is the default path and needs no key; a client that cannot do OAuth uses a personal API key created with the MCP Server preset, sent as an Authorization: Bearer header. Enterprise plans can route access through an identity provider instead of per-user keys.
One command plus a key — npx @posthog/wizard@latest mcp add, then supply credentials