Loading ToolBench
A landing page and analytics platform for DevMCP.ai that provides llms.txt generation, website scanning, AI-powered analysis, visitor tracking, and metrics for understanding AI agent adoption of APIs
This MCP server exhibits significant gaps in definition quality across multiple dimensions. While tool names generally follow verb_noun conventions, most descriptions are either missing implementation details or lack clarity about when to use each tool. Input schemas are incomplete, many parameters lack type specifications in visible code. Output schemas are not documented. Error handling is minimal or absent. The 'brilliant-recap' and 'brilliant-telemetry' tools present security and composition concerns (combining multiple responsibilities). Most critically, the server conflates two separate domains: website scanning/llms.txt generation (8 tools) and visitor analytics/email marketing (4 additional tools inferred from code). This mixing of concerns violates single-responsibility principle. Only 3 tools have verifiable input schemas with complete type information from source code.
Analyzes website traffic data from Supabase and generates AI-powered insights using Claude about visitor behavior, acquisition sources, engagement patterns, and conversion signals
Evaluates an llms.txt file or checks if a domain has llms.txt, scoring clarity, completeness, AI readability, examples, authentication documentation, and endpoint coverage. Provides recommendations for improvement
AI-powered llms.txt generation using Anthropic Claude. Takes scraped website data and generates a comprehensive llms.txt file that helps AI agents discover and integrate with APIs
Next.js API endpoint that scans a website and generates comprehensive llms.txt file based on website content, OpenAPI specs, GitHub repos, and documentation
Generates a personalized email recap of a visitor's session using Claude, analyzing their browsing behavior and creating relevant resource recommendations
Generates an AI-powered insight report for a single visitor session using Claude, analyzing behavior patterns and intent based on pages viewed, interactions, and location data
Returns aggregated metrics dashboard with overview, discovery timeline, AI agent statistics, and actionable insights about AI agent adoption and API success rates
Multiple tools combine unrelated responsibilities. 'brilliant-recap' sends emails AND generates AI summaries AND notifies Slack. 'brilliant-telemetry' tracks sessions AND generates insights AND posts to Slack. Each should be split into separate, single-purpose tools.
Input schemas for 'analyze-traffic', 'brilliant-recap', 'brilliant-telemetry', 'metrics' are incomplete. Many parameters lack explicit type definitions (string, number, object, array) in visible source code. LLMs cannot infer parameter types from names alone.
No output/response schemas documented for any tool. LLMs cannot determine what fields to expect from these tools, preventing proper downstream tool chaining and data extraction.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 6 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 39 | - | v1 |
Retrieves aggregated metrics for a domain including total discoveries, unique AI agents, discovery trend over time, AI agent breakdown percentages, and implementation rate
Comprehensive website scanning endpoint that scrapes a website for technical and marketing information, optionally generates llms.txt with AI
Handles demo request form submission, sends email via Resend and generates personalized AI response based on visitor context and company domain
Tracks API usage events when AI agents call endpoints documented in llms.txt, recording success/failure, response times, and error codes
Tracks when an AI agent discovers and accesses an llms.txt file via beacon pixel, identifying the agent type and recording the discovery event
'brilliant-recap' and 'brilliant-telemetry' accept plain 'pages_viewed' and 'clicks' arrays without documenting the expected object structure (what fields are required in each array element?). LLMs must guess the schema.
'demo-contact' tool accepts 'email' and validates it but the validation logic is not visible in provided source code. No error handling guidance for invalid emails or existing records.
Security risk: 'metrics' tool accepts 'apiKey' as a parameter. Per MCP patterns, credentials must NEVER be tool parameters. Use server-side secret injection via environment variables. Agent traces log every parameter, keys in params leak into logs.
Descriptions are too generic or lack actionable detail. 'analyze-traffic' says it 'Analyzes website traffic telemetry' but does not explain: What telemetry sources? What kind of insights? When should I call this vs 'metrics'? Ambiguity forces LLM guessing.
Tool 'metrics' is semantically overloaded. The input parameters include 'endpoint', 'success', 'responseTime', 'errorCode', these look like request logging fields, not tool parameters. Conflates input parameters with telemetry payload structure.
No pagination or limit enforcement documented. Tools like 'analyze-traffic' and 'metrics' may return large datasets. No mention of max_results, page_size, or cursor-based pagination. Large result sets waste context and invite hallucination.
No error handling guidance. No tool description explains what to do if an operation fails: is it retryable? Should the user be asked? Is it fatal? LLMs have no recovery path.
Composition weakness: 'scan' generates llms.txt, and 'evaluator' evaluates llms.txt. No documentation on whether evaluator accepts the output of scan directly or requires a different format. Broken chaining forces discovery detours.