MCP server for the true402 machine-native marketplace — pay-per-call AI + web + Base on-chain tools over x402 (USDC on Base): LLM inference, SEO/GEO audit, web extract, link preview, robots/AI-crawler check, security headers, and on-chain DeFi trading signals (token rug/honeypot safety, new token pairs, liquidity-pull/rug alerts, whale swaps).
The server exposes 11 tools with mixed quality. Most tools have adequate descriptions (150-350 chars), but parameter schemas vary significantly. 9 of 11 tools have reasonable input schemas with type definitions, but descriptions are inconsistent across parameters. The 'chat' tool has a well-structured schema with proper type annotations and descriptions. However, several tools (liquidity_pulls, whale_swaps) have empty input schemas {}, which violates the parameter documentation requirement. Tool naming follows verb_noun convention reasonably well (list_models, web_extract, token_safety), but descriptions sometimes conflate multiple concerns (e.g., seo_audit mixes SEO and GEO scoring in a single tool description without clarity on when each is needed). Output schemas are NOT documented anywhere in the source code, we cannot see what fields these tools return, which is a critical gap for composition and agent planning. Error handling descriptions are minimal; tools mention 'PAID x402 service' but do not explain what happens on payment failure, wallet insufficiency, or network errors. The tool descriptions are optimized for human reading but lack LLM-specific guidance on when to select a tool over similar ones.
Send a chat completion request to an LLM via true402 (PAID x402 service, USDC on Base). Requires a funded wallet (WALLET_PRIVATE_KEY) on the MCP server.
Analyse a URL's HTTP security headers (HSTS, CSP, X-Frame-Options, and more) into a present/missing breakdown plus a 0-100 score, along with status, HTTPS flag, and server banner. PAID x402 service (USDC on Base) — needs a funded wallet on the MCP server.
Fetch a URL and return its link-preview / Open Graph unfurl card: title, description, image, siteName, type, canonical URL, favicon, and theme color (image/canonical/favicon resolved to absolute URLs). PAID x402 service (USDC on Base) — needs a funded wallet on the MCP server.
Liquidity-pull / rug alerts on Base — liquidity-removal (Burn) events on recently-launched DEX pools (Uniswap V3 + Aerodrome). Returns the pool, token, and WETH/USDC amount removed (the rug magnitude), newest first — an early rug warning. Cross-check the token with token_safety.
List every LLM model available through true402's pay-per-call inference, with live per-token pricing (3% over provider cost, min $0.0001/request). Returns each model's id, provider, and input/output token prices across OpenAI, Anthropic, Google, Groq, Mistral and Together — so an agent can pick a model and know the exact cost before paying via x402 (USDC on Base, no account or API key).
Two tools (liquidity_pulls, whale_swaps) have empty input schemas {}. Per hard scoring rules, schema score must be 0 for tools without proper input definitions. This prevents LLMs from understanding whether parameters are required, typed, or what constraints apply. Both tools should declare their filtering/pagination parameters if they exist.
No output schemas documented anywhere in the codebase. Tools return structured JSON (e.g., seo_audit returns 'a structured JSON report: meta tags, an SEO score...'), but the actual field names, types, and nesting are not documented. Agents cannot plan downstream tool calls or extract data reliably without knowing response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Recently-created Base DEX pairs (Uniswap V3 + Aerodrome) — fresh token launches for trading/sniper agents. Returns each new token, its quote (WETH/USDC), pool, fee|stable, block and approx age, newest first. Bundle with token_safety for a pre-trade rug/honeypot check. On-chain log indexing, no API key. PAID x402 service (USDC on Base) — needs a funded wallet on the MCP server.
Report a site's AI-crawler policy (GPTBot, ClaudeBot, Google-Extended, PerplexityBot, and ~15 other AI bots: allow | block | unspecified), plus declared sitemaps and whether an llms.txt is present. Pass any URL on the target site. PAID x402 service (USDC on Base) — needs a funded wallet on the MCP server.
Audit a web page for SEO + GEO (generative-engine-optimization). Returns a structured JSON report: meta tags, an SEO score with per-category breakdown + issues, a GEO score with breakdown + issues, and a combined percentage. PAID x402 service (USDC on Base) — needs a funded wallet on the MCP server.
Rug/honeypot safety check for an ERC-20 token on Base (on-chain reads, no API key): ERC-20 conformance, ownership renounce, mint capability, WETH/USDC liquidity depth (Uniswap V3 + Aerodrome), and a buy/sell honeypot simulation (a gas-free eth_call that round-trips a tiny WETH→token→WETH trade to catch tokens you can buy but not sell). Returns a 0-100 score + risk band + flags. PAID x402 service (USDC on Base) — needs a funded wallet on the MCP server.
Fetch a URL and return its clean readable text, a light markdown rendering, all links (in document order), and metadata (title, description, word count, byte size). PAID x402 service (USDC on Base) — needs a funded wallet on the MCP server.
Whale swap alerts — large swaps on Base DEX pools (Uniswap V3 + Aerodrome) by whale wallets. Returns the pool, token(s), swap direction, amounts, and whale address, newest first. Use to detect front-runs and price slippages.
Descriptions lack error recovery guidance. Tools mention they are 'PAID x402 service' but do not explain: What happens if wallet is unfunded? What happens on payment failure? Are calls retryable? Should the agent ask the user for more funds? Per pattern:recovery-guide, errors must guide the LLM to the next action.
seo_audit tool description conflates two concerns (SEO audit + GEO audit) without clarity on when each is needed. The tool accepts a 'mode' enum (both/seo/geo) but the main description does not explain the difference between SEO and GEO scoring or when an agent should pick one over both. Per pattern:tool, each tool should have one clear responsibility, or describe multi-mode behavior clearly.
Parameter descriptions sometimes lack format/constraint guidance. Examples: 'token' parameter in token_safety says 'ERC-20 contract address (0x…)' but does not state if it must be checksummed, the minimum length, or what happens if an invalid address is passed. Per pattern:constrained-input, all format expectations should be explicit.
Tool descriptions do not mention pagination or result limits. Tools like new_pairs accept a 'limit' parameter (1 - 200, default 50) but do not explain what happens if the agent requests more items than exist, whether results are ordered, or if there is a cursor for continuation. Per pattern:paginated-result, pagination expectations must be explicit.