> AI agents: this is one page from PostHog's docs. Full index of Markdown docs for LLMs: https://posthog.com/llms.txt

# Use endpoints in PostHog Desktop

**Open beta**

Both Endpoints and PostHog Desktop are in beta.

[PostHog Desktop](/docs/posthog-desktop.md) is where endpoints meet the code that calls them. The agent has your repo, your endpoints, and PostHog's [endpoint skills](/docs/posthog-desktop/skills.md) in one place, so it can build an endpoint and wire it into your app in the same session.

## Before you start

1.  **Create an endpoint** – [define your data and publish it](/docs/endpoints/start-here.md) so there's an endpoint to call.
2.  **Install PostHog Desktop** – [download the app](/docs/posthog-desktop/download-posthog-desktop.md) and connect it to your project.

## What you can do here

-   **Build and consume an endpoint in one pass** – the agent creates the endpoint, tests it, then writes the typed client code that calls it.
-   **Audit what you have** – find endpoints nothing calls anymore, or versions still pinned by deployed clients.
-   **Diagnose slow endpoints** – work through execution logs and materialization status to find where the time goes.
-   **Copy endpoints between projects** – recreate the same query in staging or another project.
-   **Read the Self-driving inbox** – endpoint failures land there as signals, and the inbox is available in Desktop like everywhere else.

The agent reaches PostHog through the same [MCP tools](/docs/endpoints/surfaces/mcp.md) it would use anywhere else, so anything you can do over MCP you can do here.

## What endpoint failures send to Self-driving

Endpoints raises a [signal](/docs/self-driving/signals.md) in two situations:

-   **A run fails.** The signal carries the endpoint name, the version, the executed query, whether it ran materialized or inline, and the error – so whatever picks it up can reason about the cause rather than guess at it.
-   **A breakdown hits its limit.** When an insight-based endpoint returns an "Other" bucket, callers are silently seeing aggregated values for the long tail. The signal flags it and suggests a lower-cardinality breakdown or a tighter filter.

There's no endpoints-specific scout investigating these. They join the same queue as every other [signal source](/docs/self-driving/inbox/sources.md): the inbox deduplicates them, groups related ones, and turns them into a [report](/docs/self-driving/reports.md) you read alongside the rest of your work. An endpoint failure doesn't open a pull request on its own – if the report points at a code fix, you're the one who starts it.

## Related

-   Read the reports in the [Self-driving inbox](/docs/self-driving/inbox.md).
-   Do the same actions from any MCP client over [MCP](/docs/endpoints/surfaces/mcp.md).
-   Understand what the log lines mean in [usage and logs](/docs/endpoints/usage-analytics.md).

### Still have questions?

Ask PostHog AI

### Was this page useful?

HelpfulCould be better