Run SQL and manage views with MCP

The PostHog MCP server lets your AI coding agent run HogQL queries, manage Data Warehouse views, and explore your schema directly from your code editor. Query events, create and materialize views, and work with warehouse tables – without switching to the PostHog app.

This works in any MCP client – Cursor, Windsurf, Claude Code, VS Code, and others.

How it works

With MCP, your coding agent can:

Run ad-hoc queries – "How many purchase_completed events happened this week?" to quickly check data
Explore your schema – "What tables do I have?" or "Show me the columns in stripe_charges" to discover available data
Generate queries from questions – "Write a HogQL query to find the top users by event count" to get a query you can refine
Query across sources – Join PostHog events with Stripe, HubSpot, or any linked warehouse table
Manage views – Create, update, materialize, and delete saved views in your Data Warehouse programmatically

SQL tools

The MCP server provides these tools for running SQL:

Tool	Description
`query-run`	Run a HogQL query and return results. Supports any valid HogQL syntax including joins, CTEs, subqueries, and window functions.
`query-generate-hogql-from-question`	Generate a HogQL query from a natural language question. Returns the query string so you can review or modify it before running. Requires AI data processing approval.
`property-definitions`	List all event and person properties in your project. Useful for understanding what data is available before writing a query.
`event-definitions-list`	List all event definitions in your project with volume and usage data.

View management tools

The MCP server also provides tools for managing Data Warehouse views:

Tool	Description
`view-list`	List all saved views in your project with their materialization status, sync frequency, and column schema.
`view-get`	Get full details of a specific view by ID, including its HogQL query and run history.
`view-create`	Create a new view from a HogQL query. Upserts by name if a view with the same name exists.
`view-update`	Update a view's name, HogQL query, or sync frequency.
`view-delete`	Delete a view (soft delete).
`view-materialize`	Enable materialization for a view with a 24-hour refresh schedule.
`view-unmaterialize`	Revert a materialized view back to a virtual query.
`view-run`	Trigger a manual refresh of a materialized view.
`view-run-history`	Get the last 5 materialization run statuses for a view.

Example prompts

Try these with your MCP-enabled agent:

Run a HogQL query to find the top 10 events by count today.
What tables are available in my data warehouse?
Write a query joining events with stripe_subscriptions to find revenue per user.
Generate a HogQL query to calculate weekly active users by cohort.
Show me 5 sample rows from the hubspot_contacts table.
What properties are available on the purchase_completed event?

For querying alerts:

Query system.alerts to find all enabled alerts in the project.
Show me alerts that check hourly and their last check results.
Write a HogQL query to find alerts that fired in the last week.

For view management:

List all my saved views in the data warehouse.
Create a view called 'daily_active_users' that counts distinct user_ids per day.
Materialize my 'revenue_summary' view for faster queries.
Trigger a refresh of my 'user_metrics' materialized view.
Show me the run history for my 'sales_data' view.

Install the MCP server

The recommended way to install is with the AI wizard:

Terminal

npx @posthog/wizard mcp add

The wizard supports Claude, Cursor, Windsurf, VS Code, and more. You can also configure it manually.

See the MCP server docs for full setup instructions.