Loading ToolBench
Web and AI agent analytics platform exposed as an MCP server, providing tools to create accounts, track events, and read analytics reports
Strong foundation with well-structured tool definitions, comprehensive parameter descriptions, and proper schema validation. All 5 tools have clear names starting with action verbs (create_, track_, get_). Descriptions are detailed and contextual (avg ~180 chars), exceeding the 10-1024 char baseline. Input schemas are complete with type definitions and enums where appropriate. However, output schemas are not documented, LLMs cannot predict response structure. Error handling guidance is absent. Security considerations around API key handling are mentioned in code comments but not reflected in tool descriptions. Tool annotations present but incomplete (missing some hints).
Create a free analytics account, no card required and no signup form, for web analytics and AI agent analytics. Returns the account ID used to track events and install the website tag, an API key for reading reports through this server, and a one-time claim link to the dashboard (or emails a sign-in link when `email` is given). Create one account per project and reuse it — never one per run.
Which account the API key belongs to, and the projects, website domains and traffic channels reporting into it — the values the other reports' project and domain filters take.
Website and app traffic summary: sessions, unique visitors, identified users, pageviews, bounce rate, average engagement time, scroll depth, clicks and outbound links.
How to add analytics to a project permanently: the one-line website tracking script (a free Google Analytics alternative), the event API URL for AI agents, CLIs and backends, and the config to connect this MCP server to Claude, Cursor, VS Code or other MCP clients.
Record one analytics event: a pageview, a tool call, a signup, a purchase, an error — anything that happened in a website, app or AI agent. Numeric attributes are summed and charted as metrics (revenue, tokens, sats); `user` attributes the event to a person or account for per-user analytics. Events sharing a session_id form one session with page flow. Send the amount for this one event, never a running total, and never credentials, prompts or completions.
Output schemas not documented. LLMs cannot predict response structure for any tool, forcing them to reason about unknown fields and risking context loss on chained calls.
Error handling guidance absent. Tools provide no recovery hints (e.g., 'If account not found, call create_analytics_account'). Agents cannot self-correct on failures.
get_account has minimal description (70 chars). Lacks context on when to call it vs other tools and what structure it reveals.
API key security mentioned in code comments but not surfaced in tool descriptions. Agents need explicit guidance that read tools require Authorization header and write tools do not.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | A | 81 | 2026-07-28+ | v2 |
Tool annotations incomplete. readOnlyHint and destructiveHint present, but idempotentHint and openWorldHint are static across all tools. track_event is non-idempotent (duplicate events recorded); this should be explicitly marked.