Loading ToolBench
HTTP 402 Multi-Tier Agent Intelligence Suite: Scraper ($0.005), Edge LLM Digest ($0.025), Security Audit ($0.080), Deep Search ($0.050), and Twitter/X Intelligence ($0.050) on Base L2
The server presents 6 tools with reasonable naming (verb_noun convention) and non-trivial descriptions, but exhibits several critical gaps in schema documentation and parameter specification that prevent higher scoring. Tool names are clear and action-oriented (clean_web_scrape, synthesize_web_digest, audit_web_signal, clean_web_search, twitter_search, twitter_profile_lookup). Descriptions range 109-189 characters and reference the underlying business logic (HTTP 402 microtransactions, LLM synthesis, security audit). However, input schemas are minimal: most parameters lack detailed constraint documentation, type specifications exist but descriptions are sparse, and output schemas are completely undocumented. No error handling guidance visible. Parameters accepting arbitrary strings (url, query, username) lack format constraints, enums, or validation rules. The server scores above the 50th percentile due to clear naming and basic parameter presence, but below 70 due to incomplete schema specification and absence of output documentation.
Performs security, credibility, and phishing risk analysis for any website or smart contract landing page. Returns credibility score (0-100) and risk signals. Automatically handles HTTP 402 microtransactions (0.080 USDC on Base).
Scrapes any public webpage and returns clean, sanitized, token-efficient Markdown for LLM ingestion. Automatically handles HTTP 402 microtransactions (0.005 USDC on Base).
Performs deep web research across multiple sources, extracting clean Markdown summaries from top results. Automatically handles HTTP 402 microtransactions (0.05 USDC on Base).
Scrapes target webpage and executes Edge LLM (Llama 3) context synthesis to extract executive summary, key takeaways, and structured entities. Automatically handles HTTP 402 microtransactions (0.025 USDC on Base).
Extracts public Twitter/X profile bio and recent tweets for any username without the $100/mo API fee. Automatically handles HTTP 402 microtransactions (0.03 USDC on Base).
Output schemas completely undocumented. No specification of what fields each tool returns, data types, structure, or pagination info. LLMs cannot plan downstream calls or extract required data without explicit return type documentation.
String parameters (url, query, username) accept arbitrary input with no format constraints, enums, length limits, or validation rules. Descriptions lack actionable constraints (e.g., 'must start with http:// or https://' in mcp.json for clean_web_scrape is good, but others lack this). LLMs cannot predict valid inputs.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 56 | 2026-07-28+ | v2 |
Searches public tweets, cashtags ($BTC, $BASE), and sentiment across Twitter/X without the $100/mo API fee. Automatically handles HTTP 402 microtransactions (0.05 USDC on Base).
No error handling guidance visible in tool definitions. When a URL is unreachable, a Twitter user doesn't exist, or a credential fails, what should the LLM do? No recovery paths, retry hints, or error classification documented.
Numeric parameters (limit) lack min/max constraints. E.g., clean_web_search 'limit' and twitter_search 'limit' have no bounds, LLM could pass 999999 or negative values. No description states allowed range.
Parameter descriptions are minimal (16-39 chars) and lack detail on data format, expected input structure, or examples of valid input. 'The webpage URL to synthesize' tells the LLM almost nothing about format, encoding, or constraints.
No documentation of pagination, result limits, or result structure. clean_web_search and twitter_search both take 'limit' but no output schema specifies: are results an array? Is there a total count? A next_cursor? Can results exceed the limit?