Analytics

Let agents explore PostHog before they ship a change

PostHog can answer a product question and change the product surface that produced it. AIP keeps those powers distinct: agents may query trends, funnels, errors, and experiments while flag, experiment, or survey changes remain confirmable operations.

Updated 1 September 2026 13 highlighted tools 9 read · 4 write

What your agent can do

The read highlights use current PostHog MCP names and read-only annotations. Flag, experiment, and survey creation or updates are writes. AIP can Hide, Confirm, or Allow each tool after the PostHog account and project are chosen.

Tool What it does Access
read-data-schema Discover events, actions, properties, and values before constructing an analytics query. read
query-trends Analyze one or more product metrics over time with filters, breakdowns, and formulas. read
query-funnel Measure conversion and drop-off through an ordered sequence of events or actions. read
insight-query Execute a saved insight and return its actual results without changing the saved definition. read
query-error-tracking-issues-list List and filter error-tracking issues with occurrence, user, and session impact. read
feature-flag-get-all List feature flags in the active PostHog project. read
create-feature-flag Create a feature flag definition and rollout conditions. write
update-feature-flag Change an existing flag's definition, targeting, variants, or rollout state. write
experiment-list List experiments in the active project without changing them. read
experiment-results-get Run an experiment's metric queries and return results with exposure data. read
experiment-create Create an experiment definition; launching remains a separate operation. write
surveys-get-all Search and filter surveys in the active project. read
survey-create Create an in-app survey or hosted form, preferably as an unlaunched draft. write

PostHog's current MCP inventory contains far more operations across analytics, error tracking, flags, experiments, surveys, logs, dashboards, and other products. These verified highlights are intentionally not a total.

Capability snapshot: 1 September 2026 · provider source

Ask the product before changing the product

PostHog’s MCP server spans both observation and action. An analyst agent can discover the schema, run trends and funnels, inspect errors, and read experiment results. A growth agent might also prepare a flag, experiment, or survey. Treating those as one permission would make a simple product question carry unnecessary rollout authority.

In the AIP control room, query tools can stay available while creation and update tools require confirmation. The integration catalog then lets the same agent correlate PostHog evidence with Sentry incidents or Linear delivery work without sharing provider credentials between clients.

Connect PostHog to an AIP agent

  1. Select PostHog in the AIP dashboard, authenticate through PostHog OAuth, and choose the organization and project context the connection should use.
  2. Review the live discovered inventory. Start with schema and query tools; add flag, experiment, dashboard, survey, or error-triage writes only for a role that owns them.
  3. Give the generated AIP URL to Claude, Codex, Cursor, or another client. AIP keeps the PostHog token and exposes only that agent’s permitted surface.

PostHog’s inventory is intentionally broad and continues to evolve. AIP’s MCP/OpenAPI gateway avoids copying that moving configuration into every client, while the dated table above remains a set of concrete examples rather than an exhaustive contract.

Keep analysis reproducible

Ask an agent to state the event names, property filters, date range, and aggregation it used. For a proposed flag or experiment, require the current definition, intended audience, metric, rollout, and rollback condition before approval. That turns a natural-language answer into something a teammate can check in PostHog.

Separate product areas may deserve separate project connections or agents. Use the pricing page to plan that boundary; do not grant a global PostHog surface merely because a single cross-project question is convenient.

Tool policy does not replace PostHog's OAuth grant PostHog decides the organizations, projects, regions, products, and scopes available to the connection. AIP's per-tool policy controls what an agent can see and execute through AIP; it does not broaden or rewrite the provider-side OAuth grant, and the live inventory can change upstream.
Separate PostHog analysis from rollout authority
Let agents query the product freely while flags, experiments, and surveys stay supervised.
Connect an account

What to ask your agent

Product questions improve when the prompt defines an event, audience, and comparison period. Mutation prompts should also state the rollout or launch boundary and ask for review first:

> Chart weekly activation for the last quarter and break it down by acquisition channel.
> Build a funnel from signup to first successful agent run, then identify the largest drop-off.
> Which new error-tracking issues affected the most users after the latest release?
> Show the current checkout-redesign flag definition and estimate who its conditions include; do not edit it.
> Draft an experiment for the onboarding CTA with primary and guardrail metrics, but leave it unlaunched.
> Create a three-question churn survey as a draft and show the targeting conditions before launch.

Other integrations