Loading ToolBench
A Model Context Protocol server with Exa for web search, academic paper search, and Twitter/X.com search. Provides real-time web searches with configurable tool selection, allowing users to enable or disable specific search capabilities. Supports customizable result counts, live crawling options, and returns content from the most relevant websites.
Server provides 7 tools with consistent naming (verb_noun pattern: web_search_exa, research_paper_search, etc.) and detailed descriptions (150-300 chars, well above 10-char minimum). All tools are READ_ONLY, reducing security risk. However, schemas are inferred from parameter documentation rather than explicitly visible in code, only parameter names and descriptions are shown, not formal JSON Schema type definitions. Input validation and error handling are not visible in the provided source. Output schemas are undocumented. Tool descriptions are strong but could be more concise (some exceed 300 chars). No tool annotations (readOnlyHint, etc.) detected.
Research companies using Exa AI - performs targeted searches of company websites to gather comprehensive information about businesses. Returns detailed information from company websites including about pages, pricing, FAQs, blogs, and other relevant content. Specify the company URL and optionally target specific sections of their website.
Find competitors of a company using Exa AI - performs targeted searches to identify businesses that offer similar products or services. Describe what the company does (without mentioning its name) and optionally provide the company's website to exclude it from results.
Extract content from specific URLs using Exa AI - performs targeted crawling of web pages to retrieve their full content. Useful for reading articles, PDFs, or any web page when you have the exact URL. Returns the complete text content of the specified URL.
Search LinkedIn for companies using Exa AI. Simply include company URL, or company name, with 'company page' appended in your query.
Search across 100M+ research papers with full text access using Exa AI - performs targeted academic paper searches with deep research content coverage. Returns detailed information about relevant academic papers including titles, authors, publication dates, and full text excerpts. Control the number of results and character counts returned to balance comprehensiveness with conciseness based on your task requirements.
Input schemas not explicitly visible in source code. Parameter types and constraints are documented in descriptions but not in formal JSON Schema format (no z.string().min().max(), no enums for constrained values). Cannot verify schema completeness.
Output schemas are completely undocumented. Tool descriptions do not specify what fields are returned, their types, or structure. LLMs cannot plan downstream tool calls or extract data reliably.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present. While all tools are READ_ONLY, explicit annotations would improve protocol compliance and client UX.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | <=2025-11-25 | v2 |
Search Twitter/X.com posts and accounts using Exa AI - performs targeted searches of Twitter (X.com) content including tweets, profiles, and discussions. Returns relevant tweets, profile information, and conversation threads based on your query. You can search for a user by x.com/username or from:username
Search the web using Exa AI - performs real-time web searches and can scrape content from specific URLs. Supports configurable result counts and returns the content from the most relevant websites.
Error handling and recovery guidance not visible in source. No evidence of actionable error messages, error classification (retryable vs fatal), or recovery suggestions.
Tool descriptions are verbose (some 300+ chars). Descriptions should be 10-200 chars for optimal LLM parsing. Current length wastes tokens and buries key details.