Loading ToolBench
High-performance Multi-MCP server optimized for concurrent requests with SSE (Server-Sent Events) transport, supporting multiple integrated MCP servers (Resend, Tavily, Firecrawl, Notion, Apify, Apollo, BuiltWith)
This multi-server aggregator exposes 15 tools across 5 backend services (Resend, Tavily, Apollo, BuiltWith). Quality is inconsistent. Most tools have basic descriptions and input schemas, but lack proper output schema documentation, error handling guidance, and some parameter descriptions are minimal. The server loads adapters dynamically from separate modules, so tool definitions are distributed across multiple files rather than centrally visible, this makes verification incomplete. Most tools fall into the 50-65 range (fair to poor), with a few like tavily-search reaching 70 due to comprehensive parameter documentation. No tool achieves 80+. Average across 15 tools: 58.
Get employees of a specific company using Apollo.io API
Get email address for a person using Apollo.io API
Enrich organization data using Apollo.io API
Get job postings for an organization using Apollo.io API
Search for organizations using Apollo.io API
Enrich person data using Apollo.io API
No output schemas documented for any tool. Agents cannot know what fields to extract from responses, cannot plan downstream chaining, and waste tokens parsing unstructured results.
Minimal error handling guidance. No tool explains what to do if API fails, rate limit is hit, resource not found, or input is invalid. Agents receive raw failures with no recovery path.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 51 | - | v1 |
Search for people using Apollo.io API
Get comprehensive domain analysis including technologies, traffic, and more
Find related domains and relationships for a given domain
Get list of websites using specific technologies
Get detailed technology information for a website using BuiltWith API
Get technology usage trends and statistics
List all audiences from Resend. This tool is useful for getting the audience ID to help the user find the audience they want to use for other tools. If you need an audience ID, you MUST use this tool to get all available audiences and then ask the user to select the audience they want to use.
Send an email using Resend
A powerful web search tool that provides comprehensive, real-time results using Tavily's AI search engine. Returns relevant web content with customizable parameters for result count, content type, and domain filtering. Ideal for gathering current information, news, and detailed web content analysis.
send-email tool includes instructions in parameter descriptions ('You MUST ask the user for this parameter. Under no circumstance provide it yourself'). This is prompt injection risk and violates separation of concerns, agent behavior belongs in system instructions, not tool definitions. Credentials (from, cc, bcc, replyTo) should be server-side injected or configurable, not passed as parameters.
Array parameters (titles, seniorities, departments, organization_domains, organization_names, include_domains, exclude_domains) lack documented constraints. No upper limit on array length documented, can an agent pass 1000 titles? This invites resource exhaustion.
Many numeric parameters (days, max_results, employee_count_min/max, revenue_min/max, limit) lack documented valid ranges. tavily-search documents max_results 5-20 (good), but others do not. LLMs will pass invalid values without constraints.
Parameter descriptions on Apollo and BuiltWith tools are often generic or too brief (under 30 chars). Examples: 'First name of the person', 'Organization domain to enrich', 'Technology name to search for'. These lack context on format, constraints, and when to use them vs alternatives.
Tool definitions are distributed across multiple adapter files (resend, tavily, apollo, builtwith). Tool registration is not visible in the main server entry point. Inferred definitions cannot be fully verified against actual registration. This violates the 'tools must be explicitly registered' principle.
tavily-search includes example values in descriptions ('e.g., WordPress, React', 'e.g., US, GB'). LLMs often reuse example values literally instead of adapting to context. Use enums or format constraints instead.