Loading ToolBench
MCP server exposing six pay-per-call research/data tools. Pays automatically over x402 (USDC on Base) with no API keys and no subscriptions.
Seven tools with complete input schemas and descriptions. Naming follows verb_noun pattern (web_fetch, domain_intel, github_repo, token_intel, repo_diligence, web_brief, free_url_check). Descriptions are 150 - 250 chars, actionable, and include use-case guidance ('Use when your own sandbox cannot reach the open web'). All required parameters have type and description. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. No error handling guidance, no parameter constraints (enums, ranges), and no mention of rate limits or timeout behavior. Parameter descriptions lack format/constraint details (e.g., 'Absolute http(s) URL' is good, but 'maxChars' lacks min/max bounds). Tools are read-only and well-scoped, but lack idempotency hints and confirmation patterns.
Domain due diligence: RDAP registrar + creation/expiry dates + status, DNS records (A/MX/NS/TXT) and Certificate Transparency history via crt.sh. Use to vet a domain before trusting or linking to it.
Check whether a public URL is reachable and what it is: HTTP status, final URL after redirects, content type and page title. Free, no payment. Use web_fetch for full text and links.
GitHub repository facts: description, stars, forks, open issues, language, licence, last push, default branch and README excerpt.
Deep repository diligence report: release cadence, contributor spread, issue/PR health, dependency signals and a plain-language risk summary. Use before depending on, investing in, or citing a repository.
ERC-20 token facts read directly from Base mainnet: name, symbol, decimals, total supply and USDC balance of an address.
Cross-source brief: fetches several URLs and returns a consolidated summary with per-source attribution. Use for literature sweeps or competitive scans where you need citations, not raw pages.
Output schemas not documented. LLMs cannot predict response structure (fields, types, nesting) for downstream tool chaining or data extraction. Breaks the pattern:tool-chain pattern.
No error handling guidance. Errors (network timeout, invalid domain, rate limit) lack recovery hints. LLMs cannot self-correct or know whether to retry.
Parameter constraints missing. 'maxChars' and 'maxSentences' lack min/max bounds; 'url' lacks format validation hint; 'domain' lacks format example. Invites invalid LLM input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Fetch a public URL and return status, final URL, page title, cleaned text (scripts/styles stripped) and the top 20 absolute links. Use when your own sandbox cannot reach the open web.
No tool annotations (readOnlyHint, idempotentHint). LLMs cannot infer that these are safe, repeatable read-only operations. Misses current spec pattern.
No pagination or result-limit documentation. web_fetch returns 'top 20 links' but no mention of total count, next_cursor, or max result size. Breaks pattern:paginated-result for large datasets.