# signals-scout-mcp-tool-calls

signals-scout-mcp-tool-calls — a PostHog Signals skill that sifts $mcp_tool_call telemetry for MCP tools agents struggle with, then files one report per product category.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Windsurf (Devin Desktop), Claude.ai · 194 installs · ★ 101 repo stars
- Author: PostHog
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/PostHog/ai-plugin/tree/HEAD/skills/signals-scout-mcp-tool-calls/SKILL.md
- Updated: 2026-11-16
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/skills/signals-scout-mcp-tool-calls

## What it does

Inside the PostHog Signals agent (Claude sandbox), this skill queries the `$mcp_tool_call` event and looks for MCP tools agents cannot use cleanly. Error share counts only when volume and reach back it up, so failures get weighed against distinct users and sessions. Retry loops inside one session, slow p95 `$mcp_duration_ms`, oversized responses and a jump in call frequency against the previous window are findings too. Tools are bucketed by `$mcp_tool_category`, the product team that owns them.

Field coverage comes first. A precheck query over the last 7 days tells whether the event arrives at all, then the skill checks which properties the project actually sends: PostHog's own hono server carries `$mcp_error_type` and the category, external SDK servers carry the raw error text. Two reference files, `references/queries.md` and `references/metrics-dashboard.md`, hold the SQL and the dashboard recipe and are opened as needed. Every problem tool leaves with a fix hypothesis attached, written through `emit_report`; a category that already has a live report gets amended with `edit_report` rather than a second one.

Access is the gate: execute-sql, read-data-schema, write on the scratchpad and on the report channel. Without the Signals agent in the Claude sandbox nothing runs. A run that turns up nothing is a valid outcome.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add PostHog/ai-plugin --skill signals-scout-mcp-tool-calls -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/signals-scout-mcp-tool-calls/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/signals-scout-mcp-tool-calls/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add PostHog/ai-plugin --skill signals-scout-mcp-tool-calls -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/signals-scout-mcp-tool-calls/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add PostHog/ai-plugin --skill signals-scout-mcp-tool-calls -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/signals-scout-mcp-tool-calls/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install PostHog/ai-plugin skills/signals-scout-mcp-tool-calls
```

2. **Manually: put the files in place**
- File: `.github/skills/signals-scout-mcp-tool-calls/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add PostHog/ai-plugin --skill signals-scout-mcp-tool-calls -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/PostHog/ai-plugin --path skills/signals-scout-mcp-tool-calls
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/signals-scout-mcp-tool-calls/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add PostHog/ai-plugin --skill signals-scout-mcp-tool-calls -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/PostHog/ai-plugin/tree/HEAD/skills/signals-scout-mcp-tool-calls/SKILL.md

## When to use it
- spot MCP tools failing at a high rate on real traffic
- separate a genuine break from single-user noise
- explain why agents keep calling the same tool in a loop
- file a per-category report with a concrete fix
- catch a tool whose call volume jumped

## Example prompts
- Run the scout over our MCP tool calls and list the tools that fail most.
- Check `$mcp_tool_call` for the past week and file a report for the category with problem tools.
- Update the report for that category, a new tool joined the broken ones.

## Similar
- [signals-scout-web-analytics](https://affpapa.org/en/nevibecode/skills/signals-scout-web-analytics.md) — skill · 317 installs
- [signals-scout-replay-vision](https://affpapa.org/en/nevibecode/skills/signals-scout-replay-vision.md) — skill · 316 installs
- [discover-event-surfaces](https://affpapa.org/en/nevibecode/skills/discover-event-surfaces.md) — skill · 65900 installs
- [customer-churn-signals](https://affpapa.org/en/nevibecode/skills/customer-churn-signals.md) — skill · 26700 installs
- [database-scout](https://affpapa.org/en/nevibecode/skills/database-scout.md) — skill · ★ 4719 repo stars
- [logfire-query](https://affpapa.org/en/nevibecode/skills/logfire-query.md) — skill · 2200 installs

From the same set: [posthog/ai-plugin](https://affpapa.org/en/nevibecode/by/posthog/ai-plugin.md) — 82 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
