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

# WebMCP installation for MCP Analytics

If your web app registers tools with [WebMCP](https://webmachinelearning.github.io/webmcp/), [`posthog-js`](/docs/libraries/js.md) captures each tool call as a `$mcp_tool_call` event.

## Enable capture

Web

```javascript
posthog.init("<ph_project_token>", {
    api_host: "https://us.i.posthog.com",
    capture_webmcp: true,
})
```

Initialize PostHog before your app registers its tools. The SDK does not instrument tools that register earlier.

## Intent and model

By default, the SDK adds a `context` field (the agent's reason for the call) and an `llm_model` field to your tool schemas. It removes them before it calls your tool. To turn off either field:

Web

```javascript
posthog.init("<ph_project_token>", {
    api_host: "https://us.i.posthog.com",
    capture_webmcp: {
        intent: true,
        model: false,
    },
})
```

If you change these settings, register your tools again.

## Failed calls in Error Tracking

If [exception autocapture](/docs/error-tracking/installation/web.md) is enabled, a failed tool call also sends an `$exception` event. With the [slim bundle](/docs/libraries/js.md#tree-shaking), add `ErrorTrackingExtensions` next to `WebMCPExtensions`.

## Privacy

The SDK does not capture tool arguments or tool results. It redacts credentials and personal data in `context` and in error messages. Your consent, opt-out, and `before_send` settings still apply.

### Still have questions?

Ask PostHog AI

### Was this page useful?

HelpfulCould be better