Loading ToolBench
AgentGate x402 Model Context Protocol (MCP) Server — stdio bridge for machine-payable AI microservices. Exposes clean-markdown web scraping, domain intelligence, prompt injection detection, and Solana token security tools.
Server exposes 8 tools with basic schemas and descriptions, but significant quality gaps limit production readiness. All tools have input schemas with required parameters and descriptions, but lack output schema documentation, error handling guidance, and parameter validation constraints. Naming is inconsistent (agentgate_* prefix vs bare names in bridge). Descriptions are adequate (100-150 chars) but lack actionable context for LLM selection. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). No pagination support despite tools returning potentially large results. Error handling is minimal, generic 'AgentGate Gateway Error' provides no recovery guidance.
Enrich domain intelligence: DNS records, SSL validity, IP geolocation, and detected tech stack.
Deterministic firewall check for adversarial jailbreak attempts and system prompt overrides in untrusted inputs.
Fetch web pages and extract ultra-clean, noise-free Markdown. Strips headers, footers, and tracking scripts to save up to 80% LLM context window tokens. Pay-per-use via x402.
Sub-second security check on Solana tokens (LP locked status, honeypot score, freeze authority, and top holder concentration).
Sub-second anti-rug check for Solana SPL tokens (Pump.fun, Raydium). Checks LP lock status, top 10 holder concentration, mint and freeze authorities. ($0.01 USDC)
No output schema documentation. Tools return JSON responses but LLMs cannot plan downstream calls or extract required fields without knowing response structure.
Inconsistent tool naming across two implementations. index.js uses agentgate_* prefix; bridge/bin/agentgate-mcp.js uses bare names (scrape_clean_markdown vs agentgate_scrape_clean_markdown). LLMs conflate similar names, this duplication invites wrong tool selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Deterministic adversarial prompt injection, jailbreak, and system prompt override detector. ($0.002 USDC)
Instant domain intelligence: DNS, SSL, IP geolocation, and detected technology stack. ($0.01 USDC)
Ultra-fast, noise-free Markdown extraction from any web URL. 50% cheaper than Firecrawl. Strips navigation, ads, and HTML boilerplate. ($0.005 USDC)
Generic error handling. Catch-all 'AgentGate Gateway Error: {message}' provides no recovery guidance. LLM cannot determine if error is retryable, user-fixable, or fatal. No actionable next steps.
No parameter validation or constraints. 'url' and 'domain' parameters accept any string; 'mint' accepts any string. No regex patterns, length limits, or format validation. LLMs can pass invalid values (malformed URLs, non-existent domains) without early feedback.
No tool annotations. Tools lack readOnlyHint, destructiveHint, or idempotentHint. LLMs cannot determine which tools are safe to retry, which modify state, or which are read-only. Agents may retry destructive operations or skip safe retries.