Resources · Claude connector

Claude PostHog connectorWhat Claude can do in your PostHog account.

The Claude PostHog connector exposes 57 tools. 19 read insights, flags, experiments, surveys and projects, 19 create, update or delete them, and 19 carry names no source describes. Below: what Claude reads, what it can ship or remove, and what to review.

Verified Trustpilot reviews · AI, automation & growth agency

Overview

What changes once Claude can work in your PostHog

PostHog bundles product analytics, feature flags, experiments and surveys. With the connector, you stop hunting through tabs: you ask Claude how an experiment is doing or which flags are live, it reads your PostHog project, and it can also create or edit those objects when you ask. PostHog itself says the agent can read and write across its products.

Read an experiment's verdict. A product manager wants to know whether the new onboarding test is winning. Claude opens it with experiment-get and reads metric and exposure results through experiment-results-get.

Ship or clean up a feature flag. create-feature-flag, update-feature-flag and delete-feature-flag let Claude handle flags from the conversation, after feature-flag-get-all shows what exists.

Follow a survey. survey-stats gives the response statistics of one survey and surveys-global-stats the totals across all of them, so Claude can summarize feedback without an export.

What to keep in mind: 19 names on the sheet, including queries, error tracking and logs, match nothing in PostHog's tool reference, so no source describes them. Deleting tools exist for insights, dashboards, flags, experiments, surveys and actions. Nothing runs on its own either: a flag does not roll out because a metric moved, and no survey launches by itself. For events pushed into PostHog from other apps, the PostHog n8n integration covers workflows that run unattended. Different jobs; the Integrations hub lists the rest.

Vocabulary

Five words before you connect

The vocabulary of this connector, in one minute.

Connector
A link you set up once between Claude and an account you already have, so Claude can work in it while it answers you.
Tool
One named action the connector gives Claude. Claude chooses the tools it needs by itself, and the directory lists them by name.
Authorization
PostHog's sign-in step, where you hand Claude access to your account. Granted once, per person, and it can be withdrawn later.
Approval
The confirmation Claude waits for before it goes ahead with something that alters your account, shown right in the conversation.
MCP
The shared standard behind connectors. It is what lets an assistant like Claude talk to an outside service such as PostHog.
Connect

Connect PostHog to Claude in three steps

  1. 01

    Find PostHog in Claude

    Open Claude's settings, go to Customize, then Connectors, and look up PostHog. On a Team or Enterprise workspace, an Owner or Primary Owner enables the connector first, then members sign in individually.

  2. 02

    Start the connection

    Click Connect on the PostHog row and log in on the page PostHog opens itself. PostHog says its sign-in routes you to the right data region, US or EU, from your account. If the link breaks, Disconnect and reconnect.

  3. 03

    Read the authorization screen

    Check the screen before you accept. It belongs to PostHog, not to Claude, and it sets what the access covers. It may list broader rights than Claude uses day to day; asking before each action remains the default.

Tools

The 57 tools of the Claude PostHog connector

PostHog gives Claude 57 tools: 19 that read your account, 19 that change something in it, and 19 no official source describes.

Three groups: what Claude reads, what it changes, and names no official text describes. Tool names stay exactly as Claude shows them.

  • 19 read
  • 19 write
  • 19 not documented

What Claude reads (19)

19 tools

Nineteen tools that look at insights, dashboards, flags, experiments, surveys, actions, costs and projects, and leave them untouched.

insight-get

Opens one saved insight, PostHog's name for a saved analysis or chart, so Claude can tell you what it measures and how it is set up.

When it helps
a teammate shares an insight link and you want to know what it tracks before quoting it.

Sourceposthog.com · September 30, 2026 ↗

insight-query

Runs the query behind a saved insight and returns the results, so the numbers are fresh rather than whatever was cached.

When it helps
you want this week's figures from the activation insight the team already trusts.
Watch out
it runs the insight as saved; its definition stays as it is.

Sourceposthog.com · September 30, 2026 ↗

dashboard-get

Returns one dashboard, so Claude can describe which insights it holds and what the board is meant to follow.

When it helps
you inherit a growth dashboard and want a plain explanation of it before the weekly review.
Watch out
it reads the board itself; running each insight is a separate step.

Sourceposthog.com · September 30, 2026 ↗

dashboards-get-all

Lists every dashboard in the project, giving Claude the full map before it opens any single one.

When it helps
you suspect several teams built near-identical dashboards and want the full list side by side before a cleanup.
Watch out
it lists boards; their content needs dashboard-get.

Sourceposthog.com · September 30, 2026 ↗

feature-flag-get-definition

Reads the definition of one feature flag, meaning how it is set up, so Claude can explain who gets the feature and on what conditions.

When it helps
a customer sees a feature a colleague does not, and you want to check the flag's rules.

Sourceposthog.com · September 30, 2026 ↗

feature-flag-get-all

Lists all feature flags in the project, the starting point for any cleanup or audit of what is live.

When it helps
you plan a flag cleanup and want every flag in front of you before deciding which to retire.

Sourceposthog.com · September 30, 2026 ↗

experiment-get

Returns the details of one experiment, its setup and configuration, so Claude can tell you what is being tested and how it is measured.

When it helps
you join a project in the middle of a test and want to understand it before looking at any numbers.

Sourceposthog.com · September 30, 2026 ↗

experiment-results-get

Fetches an experiment's metric and exposure results, with metadata describing each metric, so Claude can read the outcome with you.

When it helps
the test has run two weeks and the team asks whether the variant is ahead.
Watch out
Claude reports the results; the decision to ship stays yours.

Sourceposthog.com · September 30, 2026 ↗

survey-get

Opens one survey by its ID to show its questions and settings, which Claude can then walk you through.

When it helps
you want to reread the wording of the churn survey before comparing its answers with another one.
Watch out
you need the survey's ID, or a search with surveys-get-all first.

Sourceposthog.com · September 30, 2026 ↗

surveys-get-all

Lists the surveys in the project, with server-side search and filters, so Claude can find the one you mean from just a few words.

When it helps
you remember a pricing survey from the spring but not its exact name or when it closed.

Sourceposthog.com · September 30, 2026 ↗

survey-stats

Returns the response statistics of one specific survey, which Claude can then summarize in plain words for you.

When it helps
the NPS survey closed yesterday and you want the headline figures before the product meeting.
Watch out
it gives statistics for one survey; the cross-survey view is another tool.

Sourceposthog.com · September 30, 2026 ↗

surveys-global-stats

Returns aggregated response statistics across every survey in the project, a bird's-eye view of how much feedback comes in.

When it helps
you report to leadership on the volume of user feedback collected this quarter across all your surveys.
Watch out
totals only; one survey's detail is in survey-stats.

Sourceposthog.com · September 30, 2026 ↗

action-get

Opens one action, PostHog's way of grouping events under a single name, to show exactly how it is defined.

When it helps
the signup action seems to count too much and you want to see which events it groups together.

Sourceposthog.com · September 30, 2026 ↗

actions-get-all

Lists all the actions defined in the project, so Claude knows which named behaviors already exist before building anything on top of them.

When it helps
you want to avoid defining a checkout action that someone already set up last year.

Sourceposthog.com · September 30, 2026 ↗

get-llm-total-costs-for-project

Fetches the total daily LLM costs of a project for each model over a number of days you choose.

When it helps
the finance lead asks what your AI features cost per model over the last month.
Watch out
it reports costs PostHog tracks; it does not cap them.

Sourceposthog.com · September 30, 2026 ↗

read-data-schema

Explores your events, actions, properties and property values, so Claude learns how your data is structured before answering questions about it.

When it helps
you are new to the project and want to know which events exist before asking for any metric.

Sourceposthog.com · September 30, 2026 ↗

organizations-get

Fetches the PostHog organizations you have access to, which tells Claude where it is allowed to look.

When it helps
you belong to a client organization and to your own, and want to confirm which one Claude is reading before it answers.

Sourceposthog.com · September 30, 2026 ↗

projects-get

Fetches the projects you can reach in the current organization, so Claude picks the right one before running any query.

When it helps
production and staging live in separate projects and you want numbers from production only, never from test traffic.

Sourceposthog.com · September 30, 2026 ↗

What Claude changes (19)

19 tools

Nineteen tools that create, alter or remove PostHog objects. No source gives them a confirmation of their own, so the general rule below applies.

insight-update

Approval: see the rule

Changes a saved insight, for instance its name, description or settings.

What Claude asks for
no source describes a confirmation for this tool, so the general approval rule below applies.
When it helps
the insight's title no longer matches what it measures and confuses new colleagues.

Sourceposthog.com · September 30, 2026 ↗

insight-delete

Approval: see the rule

Removes a saved insight from the project.

What Claude asks for
nothing tool-specific is documented; the general approval rule covers it.
When it helps
an old insight built on a retired event only misleads people.
Watch out
dashboards that show it lose that tile; read the prompt before accepting.

Sourceposthog.com · September 30, 2026 ↗

dashboard-create

Approval: see the rule

Sets up a new dashboard in the project, ready to hold insights.

What Claude asks for
no confirmation is described for this tool; the general rule applies.
When it helps
a new squad wants its own board for the features it owns.

Sourceposthog.com · September 30, 2026 ↗

dashboard-update

Approval: see the rule

Modifies an existing dashboard, such as its name, description or settings.

What Claude asks for
the general approval rule, since no source details this tool.
When it helps
a board built for a launch now serves as the team's permanent view and needs renaming.

Sourceposthog.com · September 30, 2026 ↗

dashboard-delete

Approval: see the rule

Deletes a dashboard from the project.

What Claude asks for
no prompt specific to this tool is published; the general approval rule below is what covers it.
When it helps
a cleanup turns up boards nobody has opened since a reorganization.
Watch out
colleagues who bookmarked it will find it gone.

Sourceposthog.com · September 30, 2026 ↗

dashboard-reorder-tiles

Approval: see the rule

Rearranges the tiles of a dashboard, the order in which its insights appear.

What Claude asks for
no tool-specific prompt is documented; the default rule applies.
When it helps
the most important metric sits at the bottom and everyone scrolls past it.

Sourceposthog.com · September 30, 2026 ↗

create-feature-flag

Approval: see the rule

Creates a new feature flag in the project, ready to gate a feature.

What Claude asks for
nothing specific is documented; the general approval rule covers it.
When it helps
engineering wants a flag for a new search feature before the code ships.

Sourceposthog.com · September 30, 2026 ↗

update-feature-flag

Approval: see the rule

Updates an existing feature flag, which alters who sees the feature it controls.

What Claude asks for
the sources are silent on a confirmation here, so it falls back on the approval rule below.
When it helps
a feature tested with staff should now reach a wider group.
Watch out
this acts on real users; check the rollout in the prompt.

Sourceposthog.com · September 30, 2026 ↗

delete-feature-flag

Approval: see the rule

Deletes a feature flag.

What Claude asks for
no confirmation is described for this tool specifically; the general rule applies.
When it helps
a feature has been fully live for months and the flag is dead weight.
Watch out
make sure the code no longer reads the flag before removing it.

Sourceposthog.com · September 30, 2026 ↗

experiment-create

Approval: see the rule

Adds a new experiment to the project in PostHog, ready to be configured and launched.

What Claude asks for
the general approval rule, as no source details this tool.
When it helps
you want to test whether a progress bar lifts onboarding completion.
Watch out
launching and reading it happen through other tools.

Sourceposthog.com · September 30, 2026 ↗

experiment-update

Approval: see the rule

Alters an existing experiment's settings.

What Claude asks for
no tool-specific confirmation is documented; the default rule applies.
When it helps
the team adds a secondary metric after the kickoff meeting.
Watch out
editing a running test can blur its results; say so in your request.

Sourceposthog.com · September 30, 2026 ↗

experiment-delete

Approval: see the rule

Deletes an experiment from the project, setup included.

What Claude asks for
no documented confirmation of its own; the approval block below sets the default.
When it helps
a test was set up by mistake and never ran, and it clutters the experiment list.

Sourceposthog.com · September 30, 2026 ↗

survey-create

Approval: see the rule

Builds a new survey in the project, which PostHog can then show to users.

What Claude asks for
nothing specific is documented; the general rule covers it.
When it helps
you want a two-question survey after checkout without building it by hand.

Sourceposthog.com · September 30, 2026 ↗

survey-update

Approval: see the rule

Rewrites an existing survey, found by its ID, such as its questions or settings.

What Claude asks for
no source describes a confirmation; the general approval rule applies.
When it helps
a question confused respondents and needs clearer wording mid-campaign.
Watch out
answers collected before the edit came from the old wording.

Sourceposthog.com · September 30, 2026 ↗

survey-delete

Approval: see the rule

Deletes a survey by its ID.

What Claude asks for
the general approval rule below, since this tool has no prompt of its own.
When it helps
a duplicate survey was created by accident.
Watch out
its answers may go with it; export what you need first.

Sourceposthog.com · September 30, 2026 ↗

action-create

Approval: see the rule

Creates an action, grouping events under one name you can then analyze.

What Claude asks for
no tool-level confirmation is described; the default rule applies.
When it helps
you want a single purchase action covering web and mobile checkout events.

Sourceposthog.com · September 30, 2026 ↗

action-update

Approval: see the rule

Redefines an existing action, which alters every insight built on it.

What Claude asks for
nothing published about its prompt, which leaves the general approval rule in charge.
When it helps
a new checkout event must join the purchase action.
Watch out
historical charts using the action will shift.

Sourceposthog.com · September 30, 2026 ↗

action-delete

Approval: see the rule

Deletes an action from the project, along with the grouping it defined.

What Claude asks for
nothing tool-specific is documented; the general rule covers it.
When it helps
two actions track the same behavior and one has to go.
Watch out
insights built on it lose their basis.

Sourceposthog.com · September 30, 2026 ↗

event-definition-update

Approval: see the rule

Updates an event definition's metadata, the descriptive layer around an event rather than the event data itself.

What Claude asks for
no source describes a confirmation here; the general approval rule applies.
When it helps
events lack descriptions and new analysts keep asking what each one means.

Sourceposthog.com · September 30, 2026 ↗

Undocumented (19)

19 tools

These nineteen names sit on the directory sheet with no matching entry in PostHog's tool reference, or with a line that does not say whether they read or alter anything.

query-run

This name appears on the directory sheet but nowhere in PostHog's tool reference, and no Claude page describes it. The page therefore says nothing about what it does or what it returns.

Watch out
whatever it does, the general approval rule applies to it.

query-generate-hogql-from-question

Listed by the directory, absent from PostHog's reference under this exact name. No source describes it, so its behavior stays unknown on this page, however clear the name may look.

Watch out
it falls under the default approval rule, like every tool.

insight-create-from-query

No tool spelled this way appears in PostHog's current reference. Without a sentence naming it, the page cannot say whether it only reads or also alters your project, and does not try.

Watch out
the general approval rule covers it.

insights-get-all

The reference documents single-insight tools, not this plural name. No official text describes this one, so the page leaves it unexplained rather than borrow the description of a neighboring tool with a similar name.

Watch out
treat it under the default approval rule.

add-insight-to-dashboard

Nothing in PostHog's reference or Claude's help pages matches this name. The directory publishes it without a description, and so does this page, on purpose, until a source says more.

Watch out
it sits under the general approval rule, with nothing more specific published.

experiment-get-all

The reference covers other experiment tools but not this one under this name. No source describes it, and the page does not stretch a name into a capability it cannot source from any official page.

Watch out
the default approval rule applies.

list-errors

PostHog's overview gives an example of asking for top errors, but no source ties that to this exact tool name. It stays undocumented on this page.

Watch out
covered by the same approval rule as the rest of the list.

error-details

No official page documents a tool with this name. Whatever it returns is not stated by any source, so nothing is claimed about it here, not even whether it only reads.

Watch out
the general approval rule below applies to it.

logs-query

PostHog mentions logs among its products, yet its reference has no entry for this name. No description exists to reformulate, so the page stops at the name and reports the gap.

Watch out
it falls under the default approval rule.

logs-list-attributes

Another log-related name with no matching line in the tool reference and no Claude page behind it. Its effect is not documented anywhere public, and the page leaves it that way.

Watch out
the general approval rule applies to it.

logs-list-attribute-values

Same situation as its neighbor: the directory lists it, no official source describes it. The page does not guess from the wording, even when the wording looks obvious at first glance.

Watch out
whatever it does, the default approval rule covers it.

event-definitions-list

PostHog's reference has an entry on event definitions under another name, but none under this one. With no sentence about it, nothing sourced can be said about its effect on your project.

Watch out
it stays under the general approval rule.

properties-list

No entry in PostHog's reference carries this name, and no Claude page mentions it. Its role is therefore not documented here, and no neighboring description stands in for it on this page.

Watch out
the default approval rule is what applies.

execute-sql

PostHog's reference says only that it executes an SQL query. No source says whether that query is limited to reading, so the page cannot place it on either side.

Watch out
read Claude's prompt carefully; the general approval rule applies.

read-data-warehouse-schema

A sibling of a documented schema tool, but this exact name has no line in PostHog's reference. Nothing official describes it, so the documented sibling's behavior is not assumed for it.

Watch out
it falls under the general approval rule.

organization-details-get

The only organization tool in PostHog's reference lists organizations, and it is not this one. No official sentence covers this name, so the page leaves it without a description.

Watch out
the default approval rule covers it, as for the whole group.

switch-organization

According to the tool reference, it sets which organization is active for the tool calls that follow. No source says whether that counts as reading or altering anything, so it stays in this group.

Watch out
no specific behavior is published; the approval block below is the reference.

switch-project

According to PostHog, it sets the active project for later tool calls. Whether that touches your data at all is not stated by any source, which is why the page does not classify it.

Watch out
the approval section below covers it by default.
Approvals

What Claude checks with you first

By default, Claude stops and asks before each action it takes on an account for someone. The prompt appears in the conversation when it matters.

No source gives any PostHog tool its own confirmation, so this default covers the whole list, deletions included. PostHog adds a warning of its own: models can be tricked into following untrusted instructions, so review each tool call before it runs. On Team and Enterprise workspaces, owners decide whether a member may skip the prompt for some actions and can restrict a connector for the whole organization. Claude also works with the rights of the person who connected and nothing wider.

Plans

Which plans include it

None of the 819 sheets in the official directory shows plan availability. The answer for one connector is published nowhere.

The published rule is general: remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile. On Team and Enterprise, an Owner or Primary Owner enables a connector for the organization before members connect. For this connector's live state, check the PostHog sheet in the official directory. On PostHog's side, connecting and calling tools costs nothing, though some tools run AI internally and may count as PostHog AI spend.

Limits

Where this connector stops

A connector is not an automation. Claude calls these tools while it answers you, so no flag rolls out and no survey launches because a metric crossed a line.

Everything here comes from PostHog's documentation and the directory sheet: no Claude help article covers this connector. The tool list is an observed floor, and PostHog's reference lists far more tools than the sheet under partly different names. The partner badge is not a security audit, and Anthropic says on every sheet that it neither picks a publisher's tools nor guarantees their behavior. Connect it knowing it comes from PostHog.

Need help

Need help connecting PostHog to Claude?

A person reads every message.

FAQ

Questions about the Claude PostHog connector

01What can Claude do with the PostHog connector?
Claude can read and manage much of your PostHog project from a conversation. Nineteen documented tools read insights, dashboards, feature flags, experiments and their results, surveys and their statistics, actions, LLM costs, your data schema, organizations and projects, and search PostHog's docs. Nineteen more create, update or delete insights, dashboards, flags, experiments, surveys, actions and event definitions. Nineteen other names on the directory sheet have no description in any official source.
02Can Claude create or delete things in PostHog?
Yes. Documented tools create dashboards, feature flags, experiments, surveys and actions, update all of those plus insights and event definitions, and delete insights, dashboards, flags, experiments, surveys and actions. PostHog itself says the connected agent can read and write across its products. Flag changes act on real users, and deletions remove objects that colleagues may still rely on, so read each request Claude shows you before you accept it, especially deletions.
03Does Claude ask before it changes anything in PostHog?
By default, yes: Claude asks for confirmation before each action it takes on an account for someone. No source describes a specific prompt for any PostHog tool, so that rule covers creations, edits and deletions alike. PostHog adds its own advice to review tool calls before they run, because a model can be misled by untrusted content. On Team and Enterprise workspaces, owners can also block writing for the whole organization.
04Which Claude plans is the PostHog connector available on?
No official source says, for one connector: none of the 819 directory sheets shows plan availability. The general rule is that remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile, and that on Team and Enterprise an Owner or Primary Owner enables a connector for the organization first. PostHog says connecting and calling its tools is free on its side, apart from AI-powered tools. Check the directory sheet for the live state.
05Does Claude see every project in our PostHog organization?
Claude works with the rights of the person who connected, nothing more. It can list the organizations and projects your account can reach, then read within them. PostHog also routes your sign-in to the US or EU region of the account you use, so sign in with the account that actually holds your data. On the Claude side, a Team or Enterprise owner can restrict what the connector may do for everyone, and members cannot undo that.
06Why does a PostHog request from Claude show up as AI spend?
Because some tools use language models internally. PostHog says connecting and calling tools is free, but those AI-powered tools may be billed as PostHog AI spend, and they only work when AI data processing is enabled in your organization settings. If a request fails for that reason, the setting is on the PostHog side. Plain reads, such as listing flags or opening a dashboard, are not described as AI-powered by PostHog.
07Claude or an automation tool for PostHog?
They serve different needs, so this is not a ranking. Claude is for questions and one-off changes in a conversation: reading an experiment's result, creating a flag, tidying dashboards, each time because you asked. An automation tool pushes data into PostHog or reacts to events on its own, for example sending signups from a form tool as events, with no one typing. Questions and judgment calls go to Claude; repeated data flows go to automation.