Use PostHog from your AI tool to investigate product data, manage experiments, and debug errors.
PostHog AI plugin connects your AI tool to PostHog and gives its agent task-specific guidance for querying analytics, managing feature flags and experiments, investigating errors, and working with warehouse data. You can ask a question in plain language, then have the agent use PostHog tools to investigate or make changes.
The agent can select relevant skills as you work; supported clients also let you choose a skill yourself. The session-capture commands are invoked manually, while hooks run on their configured events. What loads depends on your client: Claude Code includes optional session capture and usage telemetry, Codex has a permission hook, and Cursor and Antigravity install skills and the MCP connection without Claude Code's capture or write gate.
The marketplace also offers posthog-telemetry for Claude Code session capture alone. Install it instead of the main plugin when you only want session analytics; installing both runs the same upload hook twice.
Start with a funnel or chart that changed. The agent can compare segments, examine paths and retention, and find events or person properties associated with conversion or drop-off. Correlations help focus the investigation; they do not establish a cause.
Give the agent an error issue or session reference. It can inspect occurrences, rank user impact, find an informative replay, and connect errors to traces and logs when those identifiers are available.
Work through the hypothesis, variant split, rollout, exposure criteria, and metrics before launching. For a running test, investigate uneven exposures or inconsistent results and compare behavior across variants.
Inspect the source and affected schemas to distinguish credentials, schema changes, stalled jobs, and incremental-sync configuration problems. The agent can recommend the appropriate recovery action and help adjust the configuration.
Compare LLM spend by model, user, or trace, inspect failing conversations, and organize recurring failure patterns. Use those findings to design online evaluations or compare offline evaluation runs.
With access to your repository, the agent can follow framework-specific setup guidance and add events, error tracking, logs, metrics, or tracing. From a chat without code access, the cloud Wizard can connect to GitHub and propose installation through a pull request.
Build a dashboard from existing or new insights, create a PostHog canvas or Streamlit app, or schedule a report. Choose the format around how people will use the result: a chart collection, an interactive tool, or recurring delivery.
Showing all 181
Install and configure an SDK in a repository the agent can edit.
Use the cloud Wizard to propose SDK installation through a GitHub pull request.
Install and initialize the PostHog SDK for an application's framework.
Capture meaningful user actions with PostHog product analytics events.
Add feature-flag checks to control the rollout of new functionality.
Add exception capture with stack traces and source-map support.
Capture AI generations with token counts, latency, and cost information.
Add structured application-log capture and configure an OTLP exporter when needed.
Capture counters, gauges, and histograms, including existing OpenTelemetry-compatible metrics.
Add distributed tracing or send existing OpenTelemetry spans to PostHog.
Identify outdated PostHog SDK integrations and recommend upgrades.
Diagnose dropped or modified events, identity problems, and other ingestion warnings.
Explain project access using member permissions, roles, and resource-level rules.
Choose query tools and approved metric definitions before writing HogQL.
Investigate changes in product metrics through segments, paths, actors, and retention.
Find events and person properties associated with funnel conversion or drop-off.
Explore posthog-js interactions, identify selectors, and build click-based actions or analyses.
Interpret clicks, rage clicks, and scroll depth to recommend page improvements.
Inspect current visitors, recent pageviews, referrers, devices, and bot traffic.
Measure bot and crawler traffic and exclude it from analytics.
Propose, test, order, and apply rules that normalize dynamic URL paths.
Run and review the internal health check that proposes URL-cleaning rules.
Create or update dashboards by assembling existing and new insights.
Create and verify box plots from product analytics or SQL results.
Choose between a full time series and a two-point comparison.
Choose axis units for durations, currencies, ratios, and other numeric values.
Compare tracked AI citations with competing domains, search rankings, and referral traffic.
Explain PostHog usage and spend before investigating product-specific cost drivers.
Train and assess predictive models, then use daily person-level scores.
Identify external data sources needed to answer business questions in PostHog.
Connect external data, validate credentials, and select per-table sync methods.
Connect APIs without native connectors using a validated REST-source manifest.
Prioritize broken source connections, degraded syncs, and webhook-channel problems.
Diagnose a specific failing sync and recommend the appropriate recovery action.
Change sync methods, incremental fields, primary keys, CDC settings, or refresh frequency.
Choose PostHog views or external dbt and establish reusable modeling foundations.
Build reusable lookup tables and join them to events or business facts.
Model recurring revenue, churn, and customer value with consistent currency handling.
Build reusable funnel, drop-off, and time-to-convert models in HogQL or dbt.
Model retention, stickiness, and lifecycle using PostHog views or external dbt.
Define activation from early actions validated against later retention.
Maintain canonical metrics, table certifications, and reviewed relationships in the catalog.
Add and run assertions on warehouse tables, views, and catalog metrics.
Inspect failed checks and blocked view refreshes to identify corrective actions.
Find failed, stale, or unused materialized views that need attention.
Review, adjust, accept, dismiss, or assign warehouse optimization suggestions.
Plan warehouse-to-property mappings, check collisions, and run the first backfill.
Map warehouse columns onto person or group properties for targeting and analysis.
Export events, persons, sessions, or query results and download the files.
Expose a query as an endpoint with appropriate variables and freshness settings.
Connect applications to endpoints with typed clients, authentication, and variable handling.
Preview, update, deactivate, or recreate earlier endpoint queries as new versions.
Recreate endpoints in another project with collision checks and safe defaults.
Inspect endpoint runs for errors, cache misses, latency, and unexpected results.
Investigate one endpoint's latency or cost and propose query or caching changes.
Find stale endpoints, failed materializations, and unused versions before proposing cleanup.
Review experiments and flags for configuration issues, staleness, and hygiene problems.
Remove stale flag checks from code without changing the flags in PostHog.
Find recently deleted flags and establish deletion timing from activity logs.
Copy flags within an organization, disabled by default in target projects.
Resolve natural-language experiment references to IDs, asking when several match.
Create experiments from a hypothesis, rollout plan, metrics, and setup context.
Set experiment entry percentages and variant splits, including mid-run change implications.
Configure exposure criteria, metrics, multivariate handling, and interpretation of experiment results.
Manage experiment launches, pauses, exposure freezes, endings, shipping, and archival.
Check experiment exposures, setup, analysis, and mid-run changes for validity problems.
Investigate experiment support questions using project data and known diagnostic patterns.
Compare recordings across experiment variants to investigate differences in user behavior.
Create credit-sized scanners that compare recordings across an experiment's variants.
Create surveys with audience targeting, draft review, and launch-readiness checks.
Build custom survey interfaces that capture expected events and verify responses.
Diagnose survey visibility and response problems across supported SDKs.
Rank active errors by user impact and recency, then propose next actions.
Investigate one error issue through occurrences, affected users, segments, and replays.
Choose an informative recording linked to an error issue and summarize its context.
Connect error occurrences to traces, same-trace logs, and session replay.
Investigate unreadable stack frames and mismatched source maps or platform symbols.
Merge duplicate error issues or create rules for future grouping.
Create tightly scoped suppression or sampling rules; suppressed errors are permanently dropped.
Configure notifications for new, reopened, or spiking error issues.
Investigate service health and incidents through log patterns and before-and-after comparisons.
Draft log-alert thresholds, test them against history, and configure delivery.
Correlate infrastructure metric anomalies with logs and traces to investigate causes.
Investigate service latency and errors using distributed OpenTelemetry span data.
Turn a broad question into a focused shortlist of recordings.
Explain a recording using its user profile, events, and linked errors.
Investigate missing recordings through SDK signals, settings, sampling, and capture limits.
Design focused scanner questions, evidence requirements, session selection, and test batches.
Size and create Replay Vision scanners after checking credit requirements.
Summarize scanner findings, inspect recordings, and turn corroborated issues into work.
Inspect AI traces, messages, tool use, subagents, token consumption, and costs.
Investigate LLM spend, token economics, cache use, and cost regressions.
Identify high-spend AI users and investigate what drives their costs.
Compare AI-traffic clusters by behavior, cost, and latency, then inspect traces.
Inspect production traces and rank recurring AI failure patterns.
Create continuous AI evaluations from observed failures, then verify and enable them.
Inspect existing AI evaluations, test scorers, and investigate their results.
Compare offline evaluation runs and investigate case-level regressions and scorer history.
Extract AI-feature use cases from linked generation data into a Slack feed.
Turn broad MCP-usage questions into focused investigations with the appropriate tools.
Compare MCP tools by error rate, latency, and user reach.
Reconstruct a session's tool calls, apparent goal, and points of failure.
Compare agent-intent clusters, tool routing, discovery, and failure patterns.
Reconstruct users' starting goals for an MCP tool and publish a taxonomy.
Build, test, edit, enable, and monitor PostHog messaging workflows.
Create and edit workflow email templates with Liquid personalization.
Configure transactional templates and send one-person emails from application code.
Connect Supabase's Send Email Hook to PostHog transactional templates.
Preview and import Customer.io automations as draft PostHog workflows.
Build event-triggered or scheduled AI tasks with optional result notifications.
Schedule AI-written reports for delivery by email, Slack, or Microsoft Teams.
Manage scheduled insight, dashboard, and AI-report delivery, including delivery history.
Schedule private in-app reminders, optionally linked to PostHog resources.
Scout definitions run in PostHog's Signals environment and need their specified tools and scopes. Installing the plugin does not schedule them. The workflows scout writes workflow suggestions; most other scouts surface findings through Signals reports.
Browse, investigate, resolve, dismiss, snooze, or act on PostHog Inbox reports.
Rank Inbox reports using your activity, ownership, routing, and report relevance.
Search raw signal observations and inspect the evidence behind Inbox reports.
Assign monitoring work, interpret reports, reduce noise, and calibrate Signals scouts.
Inspect scout schedules, runs, memory, reports, and performance without editing them.
Create or tune Signals scouts, schedules, reporting rules, and evaluation prompts.
Find one actionable result after Self-Driving activation using existing project data.
Investigate cross-product correlations and areas outside specialist scouts' coverage.
Check frequently viewed insights and dashboards against seasonality-matched baselines.
Monitor saved product flows for conversion or engagement-rate regressions.
Monitor channel traffic, attribution, bounce rates, and landing-page failures.
Monitor page-performance percentiles and relate regressions to deploys or rollouts.
Find barely used product areas and investigate why reachable controls go unused.
Look for account-engagement patterns associated with churn risk or expansion.
Monitor account product-mix changes that may be hidden by stable totals.
Monitor revenue collection for upstream failures and configuration drift.
Monitor flag evaluations, response shifts, unused flags, and accumulated flag debt.
Monitor running experiments for exposure problems, validity threats, and lifecycle drift.
Monitor survey scores, response volume, abandonment, targeting, and recurring feedback.
Monitor recording coverage and friction patterns such as rage or dead clicks.
Monitor scanner throughput, credit constraints, score shifts, and recurring observations.
Monitor exception bursts, repeated loops, duplicate fingerprints, and issue-status regressions.
Monitor service traces for regressions and persistent performance problems.
Monitor log patterns, volume, severity, service silence, and trace-correlated bursts.
Monitor Content Security Policy violations for bursts, regressions, and suspicious domains.
Monitor AI traces for changes in cost, latency, errors, and evaluation performance.
Find widely affecting MCP failures, repeated retries, slow calls, and oversized responses.
Monitor destinations, transformations, workflows, and batch exports for delivery problems.
Monitor warehouse imports and materialized views for failures and silent staleness.
Monitor support-ticket response times, SLA breaches, backlog, and assignment concentration.
Propose email-step improvements for opted-in workflows through the suggestions API.
Monitor task-run failures and retries; keep private demand analysis in scout memory.
Check deployment timing and telemetry after merged pull requests.
Recheck resolved reports and investigate fixes that did not hold.
Review recent insight-alert firings for issues a person may have missed.
Group active project health issues and prioritize their affected scope.
Find significant activity without corresponding insights, dashboards, or alert coverage.
Check shared skills for unclear scope, broken links, excessive size, and secrets.
Create or edit freeform PostHog canvas apps through validation and publishing.
Build canvas interfaces with React and PostHog's Quill components.
Build canvas documents, graphics, and tools with HTML and browser APIs.
Arrange widgets on grid canvases and create reusable component canvases.
Load insights, queries, and connected data into canvases with traceable results.
Validate canvas sources and publish builds while handling version conflicts.
Create, deploy, operate, version, and share Streamlit apps in PostHog.
Write Python Streamlit apps for PostHog's sandbox and data-query bridge.
Investigate workflow speed, flaky CI, merge times, and stalled pull requests.
Trace a specific CI failure to its likely responsible change.
Save CI and pull-request analysis as insights, dashboards, and scheduled subscriptions.
Investigate screenshot regressions and flaky stories that block pull-request merges.
Help PostHog staff establish deployment timing from hidden annotations and merge commits.
Customize PostHog Review perspectives, validation criteria, and rules for implementing findings.
Review security, input validation, and API compatibility in PostHog Review.
Review business logic, edge cases, transformations, and queries in PostHog Review.
Review resource use, error recovery, and scaling in PostHog Review.
Check code for issues missed by other PostHog Review perspectives.
Filter PostHog Review findings for material impact before publication.
Guide safe, verifiable fixes and escalate risky PostHog Review requests.
Guide PostHog Review fixes for substantive bugs and missing functionality.
Guide PostHog Review fixes for small bugs, wording, and outdated documentation.
Discover and access shared team skills stored in PostHog.
Author and maintain shared PostHog skills, including large collections of supporting files.
Read task, artifact, and canvas feedback with its thread and version context.
Record evidenced activities and blockers from a completed PostHog task-run log.
Write actionable English with consistent vocabulary, active voice, and short sentences.
Update a mounted context wiki with durable, sourced organizational facts.
Resolve lint-report issues in the mounted context wiki before publication.
Review a context wiki for unsupported claims, contradictions, duplication, and stale priorities.
No matching results.
/posthog:llma-cc-ingestManually upload a selected Claude Code session log to LLM Analytics.
With the main plugin enabled in Claude Code, select posthog:llma-cc-ingest from the slash-command menu. Supply a session ID or JSONL path, or leave the argument empty to choose a recent session. The upload sends session data to your configured PostHog project; review your capture and privacy settings first.
See the ingest command for accepted inputs.
/posthog:llma-cc-setupConfigure opt-in Claude Code session capture for your PostHog project.
In Claude Code, select posthog:llma-cc-setup from the slash-command menu. Have your PostHog project API key and hosting region ready. Capture requires explicit enablement and a project key; privacy mode redacts prompt and output content while retaining token and cost information.
The setup command covers project-local and global settings.
/posthog:llma-cc-statusInspect capture settings and the result of the last session upload.
In Claude Code, select posthog:llma-cc-status from the slash-command menu. It checks capture settings and ~/.claude/posthog-llma-status.json for the last upload's time, event count, and result. Automatic uploads happen when a session ends, so a newly configured session may have no upload status yet.
Investigate multiple errors, identify shared patterns, and prioritize by user impact.
The agent definition is bundled for Claude Code and Cursor. Give it a time range or error set to investigate; its workflow combines issue triage, detailed investigation, pattern analysis, and recommendations.
Cursor describes plugin agents as subagents: Agent can delegate by description, or you can ask for error-analyzer by name. See Cursor's subagent guidance and the agent definition.
At Claude Code session end, uploads session events when capture is enabled.
The main plugin registers this hook when enabled in Claude Code. Uploading requires Python 3, POSTHOG_LLMA_CC_ENABLED set to true, and POSTHOG_API_KEY. Session data goes to your configured PostHog project when Claude Code exits; enable privacy mode if prompt and output content should be redacted.
Open the hooks browser in Claude Code and look for the plugin's SessionEnd entry. Check upload results with the status command listed above. To stop uploads, turn off session capture; to unload the plugin's hooks, disable the plugin in the Installed tab and apply the reload.
/hooksBefore exec calls, requests approval for selected sensitive writes and cross-project forwarding.
Enabling the main plugin registers this Claude Code PreToolUse hook. Its default prompt set covers feature-flag writes, deletions, experiment launch, shipping or reset, survey launch, and workflow enablement. Cross-project forwarding also prompts. It does not require a new prompt for every write.
Check the PreToolUse entry in Claude Code's hooks browser. To turn this gate off, set POSTHOG_MCP_EXEC_GATE_DISABLE to 1; normal client permissions still apply. The gate source documents how to change the prompt set.
/hooksAt Claude Code session end, runs the capture-only plugin's opt-in session upload.
The posthog-telemetry plugin registers this Claude Code SessionEnd hook. It needs Python 3 and the same capture settings as the main plugin. Install only one of these plugins: both register the upload hook, which otherwise processes each session twice.
Check its SessionEnd entry in Claude Code's hooks browser and inspect ~/.claude/posthog-llma-status.json after a session ends. Turn off capture to stop uploads, or disable the capture plugin in the Installed tab and apply the reload.
/hooksDuring Codex permission requests, approves known reads and leaves other calls to normal approval.
The Codex manifest selects this hook instead of the Claude Code hook set. Review and trust it at startup or in the hooks browser before it can run; changed plugin scripts require another review. The hook approves only known reads. Other calls remain subject to Codex's approval policy.
Use the hooks browser in Codex CLI to confirm its source and trust status or disable this individual hook. With approval policy never, restricted access denies calls the hook does not approve, while full access can approve calls before the hook runs. The gate's Claude Code environment settings do not change this Codex behavior.
/hooksAfter Claude invokes a bundled skill, sends a skill-usage event.
The main plugin registers this Claude Code PostToolUse hook for the Skill tool. For this plugin's skills, it sends the skill name, invocation type, whether arguments were present, and installation metadata to PostHog Cloud. It does not send the arguments themselves.
Confirm registration in Claude Code's hooks browser. To turn this telemetry off, set POSTHOG_PLUGIN_TELEMETRY_DISABLED to 1 in your shell environment or Claude Code settings. The telemetry documentation lists the fields and automatic exclusions.
/hooksWhen a PostHog skill slash command expands, sends a user-invocation event.
The main plugin's manifest registers this Claude Code UserPromptExpansion hook for PostHog slash commands. It records manual skill invocation using the same telemetry script and fields as the agent-invoked hook.
Confirm its UserPromptExpansion entry in Claude Code's hooks browser. Set POSTHOG_PLUGIN_TELEMETRY_DISABLED to 1 to stop both skill-usage telemetry hooks.
/hooksConnects the agent to PostHog analytics, flags, experiments, errors, and other project tools.
Use the PostHog marketplace instructions to choose the main plugin or session capture alone. Install one: both include the same session-upload hook.
Add the marketplace from your terminal.
claude plugin marketplace add PostHog/ai-pluginWhat do you want to install?
In Claude Code, bundled skill invocations send usage metadata to PostHog Cloud by default, including the skill name and whether arguments were present, but not the arguments. Set POSTHOG_PLUGIN_TELEMETRY_DISABLED to 1 to opt out. Session-content capture is separately opt-in.
Install the main plugin from the marketplace.
claude plugin install posthog@posthogStart Claude Code, or reload plugins in an existing session so the shell installation loads.
/reload-pluginsOpen MCP connections, select plugin:posthog:posthog, and follow the browser prompts to sign in to PostHog.
/mcpConfirm that the PostHog connection is available in the MCP view. Open the plugin browser's Installed tab to check the plugin status; typing a slash in chat also lists its available skills.
/pluginHave Python 3 available. In ~/.claude/settings.json for all projects or .claude/settings.local.json for this project, merge the capture env values into your existing settings. Replace phc_... with your PostHog project API key. The example uses EU ingestion; use https://us.i.posthog.com for a US project. Session capture includes prompt and output content unless you set POSTHOG_LLMA_PRIVACY_MODE to true.
{
"env": {
"POSTHOG_LLMA_CC_ENABLED": "true",
"POSTHOG_API_KEY": "phc_...",
"POSTHOG_HOST": "https://eu.i.posthog.com"
}
}After a Claude Code session ends, use the main plugin's posthog:llma-cc-status command from the slash-command menu to inspect the upload result. Turn off capture to stop future session uploads.
This installs posthog-telemetry: a SessionEnd hook without skills, slash commands, or an MCP server. It needs Python 3 and a PostHog project API key. Disable or uninstall the main plugin first if it is already enabled.
Install the capture plugin.
claude plugin install posthog-telemetry@posthogEnable capture in ~/.claude/settings.json for all projects, or .claude/settings.local.json for this project. Merge the env values into any existing settings. Replace phc_... with your project API key; the example uses EU ingestion, so use https://us.i.posthog.com for a US project. Capture sends session events, including prompt and output content; set POSTHOG_LLMA_PRIVACY_MODE to true to redact that content.
{
"env": {
"POSTHOG_LLMA_CC_ENABLED": "true",
"POSTHOG_API_KEY": "phc_...",
"POSTHOG_HOST": "https://eu.i.posthog.com"
}
}Start Claude Code or reload plugins in an existing session, then check for the capture plugin's SessionEnd entry in the hooks browser.
/reload-plugins/hooksThis is PostHog's direct installation method for the main plugin. Choose the marketplace method if you want the capture-only alternative.
Bundled skill invocations send usage metadata to PostHog Cloud by default, without skill arguments. Set POSTHOG_PLUGIN_TELEMETRY_DISABLED to 1 to opt out. Capturing session content is separately opt-in.
Install the plugin from your terminal.
claude plugin install posthogStart Claude Code, or reload plugins in an existing session.
/reload-pluginsOpen MCP connections, select plugin:posthog:posthog, and follow the browser prompts to sign in.
/mcpCheck the PostHog connection in the MCP view and the plugin's status in the Installed tab. Its available skills also appear when you type a slash in chat.
/pluginHave Python 3 available. In ~/.claude/settings.json for all projects or .claude/settings.local.json for this project, merge the capture env values into your existing settings. Replace phc_... with your PostHog project API key. The example uses EU ingestion; use https://us.i.posthog.com for a US project. Session capture includes prompt and output content unless you set POSTHOG_LLMA_PRIVACY_MODE to true.
{
"env": {
"POSTHOG_LLMA_CC_ENABLED": "true",
"POSTHOG_API_KEY": "phc_...",
"POSTHOG_HOST": "https://eu.i.posthog.com"
}
}After a Claude Code session ends, use the main plugin's posthog:llma-cc-status command from the slash-command menu to inspect the upload result. Turn off capture to stop future session uploads.
Follow PostHog's Antigravity instructions. This installs skills and the MCP server; Claude Code's session capture and write-command gate do not run here.
Install the plugin from your terminal.
agy plugin install https://github.com/PostHog/ai-pluginOpen MCP connections, select posthog_posthog, and choose Authenticate. Open the supplied link, sign in to PostHog, and paste the authorization code back into Antigravity.
/mcpConfirm that PostHog appears in the plugin list and check its authenticated MCP connection.
agy plugin listInstall the plugin through Claude's account plugin interface. Skills and directory plugins require a supported paid plan; enable Code execution and file creation to use skills. See Claude's plugin documentation and PostHog's Cowork instructions.
Open Customize > Plugins. Choose Add > Add marketplace, enter PostHog/ai-plugin, and sync the marketplace.
Browse the marketplace, select PostHog, and install it.
Open the plugin's connector sign-in and follow the browser prompts to connect PostHog. Installing a plugin does not by itself add an authenticated connector.
Confirm the bundled skills under Customize > Skills > Your skills and the PostHog connection under Customize > Connectors. Start a conversation with the needed skills enabled.
Use the PostHog plugin for Cursor. It includes skills and the MCP server; Claude Code's session capture and write-command gate do not run in Cursor.
Open Cursor's Customize page or marketplace panel, search for PostHog, and install the plugin at project or user level.
Follow the PostHog sign-in prompt for the MCP connection. Cursor's hosted service holds this sign-in; you may need to authenticate again after updating the plugin.
Confirm that the plugin is installed in Customize and that its skills appear when you type a slash in Agent chat. Check the PostHog MCP connection before using its tools.
Follow PostHog's Codex instructions. Plugins are supported in Codex CLI and the desktop app; the IDE extension does not support plugin installation.
Add the PostHog marketplace from your terminal.
codex plugin marketplace add PostHog/ai-pluginLaunch Codex, open its plugin browser, select PostHog, and install and enable it.
codex/pluginsStart a new session. Review and trust the plugin's permission hook when prompted, or open the hooks browser to review it.
/hooksOpen the MCP view to check that PostHog is active. On first use, follow the browser login prompt to connect your PostHog account.
/mcpConfirm that PostHog's skills appear in the available-skills list and that its MCP server is active.
/skillsThe trusted hook approves known reads; other calls follow your approval policy. With policy never, restricted access denies unapproved calls, while full access can approve calls before the hook runs. A separate [mcp_servers.posthog] entry in ~/.codex/config.toml replaces the plugin's server settings: remove that entry to use the plugin defaults, or retain it and apply the adjustments in PostHog's Codex guidance.
Use PostHog's Grok installation method for the terminal coding agent. The command includes a trust flag; review the plugin before running it.
Install the plugin from your terminal.
grok plugin install PostHog/ai-plugin --trustOn first use of a PostHog tool, follow the browser authorization prompt and sign in to PostHog.
Open Grok's extensions modal and confirm that PostHog appears among the installed plugins.
/pluginsInstall the PostHog extension from your terminal. Its MCP configuration launches a local proxy through npx, so Node.js and npx must be available.
Install the extension outside Gemini CLI's interactive session and review the requested access.
gemini extensions install https://github.com/PostHog/ai-pluginRestart Gemini CLI to load the extension. Use a trusted working folder so its stdio MCP server can connect.
On first use of PostHog, follow the proxy's browser login prompt. The bundled server runs as a local process; Gemini CLI's own OAuth support for remote HTTP and SSE servers is a different connection mode.
Confirm that PostHog appears in the installed-extension list, its skills are listed, and the MCP view shows the server as connected with tools.
/extensions list/skills list/mcpPostHog lists a Grok Bot installation through the shared Cursor Marketplace. Use Grok Bot's current Connect apps interface; PostHog's installation note calls this area Plugins.
Open Connect apps from the sidebar, search for PostHog, and add it.
Follow the authorization prompt and sign in to PostHog.
Confirm that PostHog appears under Installed and that sign-in is complete. Added means installed, not authenticated; its tools are available from your next message.