@agentutility/mcp-wordmint

Named LLM tasks. Minted at a sticker price.

Summarize, translate, classify, extract entities, generate regex, score a resume — every named text task an agent needs, with a fixed price and no prompt engineering.

Pricing: pay-per-call in USDC on Base. No subscriptions, no API keys. See per-tool prices below.

Install — Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "agentutility-wordmint": {
      "command": "npx",
      "args": ["-y", "@agentutility/mcp-wordmint"],
      "env": { "X402_PRIVATE_KEY": "0xYOUR_PRIVATE_KEY_HEX" }
    }
  }
}

Restart Claude Desktop. 46 tools appear in the tool palette.

Install — Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "agentutility-wordmint": {
      "command": "npx",
      "args": ["-y", "@agentutility/mcp-wordmint"],
      "env": { "X402_PRIVATE_KEY": "0x..." }
    }
  }
}

Funding

Send any amount of USDC on Base mainnet to the address derived from your X402_PRIVATE_KEY. The MCP server uses it to pay for tool calls automatically.

USDC on Base contract: 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913

Tools (46)

| Tool | Description | |---|---| | ai-to-human-text | (0.01 USDC/call) AI text humanizer / GPT detector bypass. Strips telltale LLM patterns (formulaic transitions, em-dash overuse, 'delve' / 'tapestry' / 'navigate the landscape'). 4 tones. | | card-resolve | (0.005 USDC/call) Card resolver / graded card string normalizer / free-form card text to canonical card object. Takes any descriptive string ('2018 Topps Update Juan Soto RC PSA 9', '1999 Pokemon Base Set Charizard Holo PSA 10', 'MTG Black Lotus Alpha BGS 8.5') and returns a structured object: sport_or_tcg, year, set, player_or_card_name, parallel, grade, grader, cert_number. Plus a confidence score and a deterministic search_key suitable for downstream cache coherence. Pure LLM extraction (Morpheus primary, Venice fallback), no upstream data lookups. The clean glue endpoint for collector-pricing bots, marketplace listing automations, AI-shopping agents, and portfolio trackers that need to normalize messy human descriptions before any further lookup. | | citation-verify | (0.01 USDC/call) Citation verifier / fact-check against URL / does-this-source-support-this-claim / hallucination detector. Fetches a source URL (HTML stripped, 100KB cap), and asks Venice claude-sonnet-4-6 whether a given claim is supported, contradicted, or absent. Returns verdict, confidence (0-1), and the strongest evidence quote when available. | | classify | (0.02 USDC/call) Zero-shot text classifier. Caller supplies labels. Multi-label mode. Calibrated confidences + reasoning. | | classify-text | (0.02 USDC/call) Text classifier / zero-shot classifier / category sorter. Caller supplies labels (2-25). Multi-label mode. Calibrated confidences + reasoning. Mistral powered. | | commit-message-from-diff | (0.01 USDC/call) Git commit message generator / Conventional Commits / AI commit-msg. Type, scope, subject, body, breaking_change flag. Ready for git commit -m. | | content-simhash | (0.001 USDC/call) SimHash / 64-bit content fingerprint / near-duplicate detection / dedup hashing / locality-sensitive hash. Pure-local 64-bit SimHash over token-level k-shingles (default k=3) using FNV-1a. Two SimHashes are 'close' (small Hamming distance) iff the underlying texts share many shingles. Returns hex + decimal forms plus token + shingle counts. Useful for content dedup pipelines, plagiarism detection, and bot-content clustering. | | cron-explain | (0.002 USDC/call) Cron expression explainer / cron parser / scheduling translator. Takes 5/6/7-field cron OR @yearly/@monthly/@daily macros. Returns plain English, frequency bucket, per-field breakdown, validation notes, runs-per-day estimate. | | cron-next | (0.003 USDC/call) Cron next fire times / cron forecaster / when does this cron run next / schedule preview / cadence calculator. Takes any 5/6/7-field cron expression (or @daily/@hourly/@weekly/@monthly/@yearly macros) plus a timezone and returns the next N firing times as ISO 8601, Unix epoch seconds, and a human-readable string in the requested timezone. Also reports whether the cadence is regular (constant interval between fires) and the interval in seconds when it is. Pure deterministic compute via cron-parser — no LLM, no upstream lookup. Companion to cron-explain which describes the cadence in English; this one tells you exactly when it fires. | | cron-parse | (0.002 USDC/call) Cron parser. Parses any 5 / 6 / 7-field cron expression or @yearly / @monthly / @weekly / @daily / @hourly macro and returns a plain-English explanation, frequency bucket, validation flag, and runs-per-day estimate. Same backend as cron-explain under a clearer slug. | | describe-image | (0.02 USDC/call) AI image descriptor / vision LLM. Modes: describe, alt_text (accessibility, ≤125 chars), OCR (extract visible text), tags (8-15 keywords), caption (single-sentence). Vision LLM powered. | | detect-language | (0.005 USDC/call) Language detector / language identification. ISO 639-1 code, full name, confidence (0-1). Mixed-language detection with secondary languages list. 100+ languages. | | detect-pii | (0.02 USDC/call) PII detector / data leak scanner. Emails, phones, SSNs, credit cards, addresses, names, IPs, API tokens. Returns matches + risk_level. Optional redaction with [TYPE] tokens. | | dictionary-define | (0.002 USDC/call) English dictionary / word definition / lookup word / pronunciation / part of speech / synonyms / etymology adjacent. Returns part-of-speech-grouped definitions with usage examples and per-sense synonyms, plus IPA phonetic and a pronunciation audio URL when available. Wraps the public Free Dictionary API — no auth, commercial-OK. | | email-draft | (0.02 USDC/call) AI email writer / cold outreach / follow-up generator. Subject + body + salutation + sign-off. 7 tones × 10 email types (cold_outreach, follow_up, decline, reply, internal_update, thank_you, intro, request, apology, general). Length controls. | | embedding-similarity | (0.002 USDC/call) Embedding similarity / cosine similarity / semantic match / vector compare / are-these-strings-similar. Embeds two strings via Venice (default model: text-embedding-bge-m3) and returns the cosine similarity as a single float in [-1, 1]. Useful for paraphrase detection, dedup, and cheap retrieval routing. | | extract | (0.02 USDC/call) Named entity extractor / NER. People, organizations, locations, dates, emails, URLs, phone numbers, monetary amounts. Mistral-powered. | | extract-entities | (0.005 USDC/call) Named entity recognition (NER) / entity extractor. Pulls people, organizations, locations, dates, emails, URLs, phone numbers, monetary amounts. Verbatim with context. Mistral powered. | | humanize | (0.01 USDC/call) AI text humanizer / GPT detector bypass. Strips LLM tells (em-dash overuse, formulaic transitions, 'delve', 'tapestry'). 4 tones. | | moderate-content | (0.02 USDC/call) Content moderation / safety classifier / OpenAI-style toxicity API. Default: harassment, hate_speech, violence, sexual_content, self_harm, spam, phishing, doxing, illegal_activity. Custom categories + verdict (allow/review/block). | | pii-redact | (0.005 USDC/call) PII redactor / mask emails phones SSNs IBANs credit cards IPs / GDPR safe text / privacy scrubber. Two-stage redaction: a deterministic regex pre-pass for structural PII (email, phone, credit card, SSN, IBAN, IPv4/v6, URL), then Venice mistral-small-3-2-24b for residual PII (full names, street addresses, dates of birth). Returns the redacted text and a list of every (type, value, masked_with) triple. | | pr-description-from-diff | (0.01 USDC/call) Pull Request description generator / PR body from unified diff / AI PR description / GitHub PR body / GitLab MR description / generate-pr-from-git-diff. Takes any unified git diff (up to 60k chars) and returns a structured PR title (<= 70 chars, imperative, conventional-commits-flavored), 2-5 summary bullets, a 3-7 item actionable test plan rendered as a Markdown checklist, plus a breaking_change boolean and migration note when applicable. Returns both the structured fields and a ready-to-paste Markdown body with ## Summary, ## Test plan, and optional ## Breaking change sections. Pairs with commit-message-from-diff at a wider scope: commits get the one-line subject + body; PRs get the full review-ready narrative. | | prompt-compress | (0.005 USDC/call) Prompt compressor / context shrinker / prompt distiller / cost-cutter for long system prompts. Rewrites a long prompt down to a target ratio of its original length while preserving every instruction, constraint, and example's intent. Drops filler words, redundant repetition, and ceremonial politeness. Powered by Venice mistral-small-3-2-24b. | | regex-from-prompt | (0.01 USDC/call) Regex generator / NL to regex / pattern builder. PCRE / JavaScript / Python / Go / RE2. Returns pattern + flags + explanation + 3-6 test examples. Live-runs JS regex on sample_text. | | regex-test | (0.002 USDC/call) Regex tester / pattern matcher / regex playground / verify-a-pattern / match-and-capture extractor. Takes a JavaScript regex (pattern + flags) and an array of test inputs, returns per-input matched / not-matched verdict, every match with byte index, all numbered capture groups, and named groups (?...). No LLM, pure JS RegExp via new RegExp(), deterministic. Catches catastrophic-backtrack-shaped patterns up front (nested unbounded quantifiers) and rejects them with a clear error. Companion to regex-from-prompt: that one generates a pattern from natural language, this one tells you whether your pattern actually matches what you think it matches. | | resume-scorer | (0.02 USDC/call) AI resume scorer / ATS keyword analyzer. Scores resume vs. job description (0-100 fit), with calibrated subscores: keyword match, experience, skills, formatting, impact. Ranked improvement suggestions. | | retrieval-rerank | (0.005 USDC/call) Retrieval reranker / RAG reranker / document scoring / top-k filter / cross-encoder substitute. Given a query and up to 30 candidate documents, scores each 0-100 for query relevance using Venice qwen3-5-35b-a3b in JSON-mode and returns sorted top-k. Useful as a second-stage reranker on top of cheap vector retrieval. | | rewrite-tone | (0.02 USDC/call) Tone rewriter / paraphraser / writing style changer. 12 tones: formal, casual, friendly, confident, empathetic, concise, playful, persuasive, apologetic, technical, simple, enthusiastic. Audience + length controls. | | semantic-chunk | (0.002 USDC/call) Semantic chunker / text splitter / RAG chunker / chunking with overlap / sentence + paragraph aware. Splits long text into chunks with three modes: 'fixed' (hard char-count windows with overlap), 'sentence' (greedy pack of sentences up to chunk_size), 'paragraph' (split on blank lines, never pack across paragraphs). Returns each chunk's text, start/end character offsets, and char count. Pure local — no upstream call. | | sentiment | (0.01 USDC/call) Sentiment + emotion analyzer. Overall sentiment + score (-1..+1) + per-emotion (joy/anger/sadness/fear/surprise/disgust). Aspect-based optional. | | sentiment-analysis | (0.01 USDC/call) Sentiment analyzer / emotion classifier / aspect-based sentiment. Overall sentiment + score (-1 to +1) + per-emotion (joy/anger/sadness/fear/surprise/disgust). Optional aspect-based scoring. | | slugify | (0.001 USDC/call) URL slug generator / slugifier / canonical-identifier maker / safe-string converter / SEO slug builder / filename slug / cache-key normalizer. Takes any string and returns a URL/identifier-safe slug in one of four modes: kebab (lower-kebab-case, default), snake (lower_snake_case), dot (lower.dot.case), or preserve_case (keep original casing). Unicode-aware: strips accents (José → jose), normalizes symbol-to-word for common cases (& → and, % → percent, @ → at, # → hash, + → plus), collapses adjacent separators, strips leading/trailing separators. Optional max_length with word-boundary-aware truncation. Pure compute, no LLM, no upstream. Cheapest endpoint in the catalog at $0.001. | | sql-from-prompt | (0.02 USDC/call) Text to SQL / NL to SQL / AI SQL generator. Postgres, MySQL, SQLite, BigQuery, Snowflake, MSSQL, DuckDB, ANSI. Optional schema. Returns SQL + explanation + tables_referenced + is_destructive + warnings. | | structured-extract | (0.01 USDC/call) Structured-data extractor / JSON-from-text / schema-guided extraction / key-value pull / form-filler. Reads free-form text and emits a JSON object conforming to a user-supplied JSON Schema. Powered by Venice zai-org-glm-4.7 (function-calling-default model) with response_format json_object. Returns the extracted object plus a validates flag against the user's schema (basic top-level type and required-key check). | | summarize | (0.01 USDC/call) AI summarizer / TLDR generator. Compress long text into tldr / bullets / paragraph / executive-summary. 30k char input. Mistral-powered. | | summarize-text | (0.01 USDC/call) AI text summarizer / TLDR generator. 4 styles: tldr (1-2 sentences), bullets (3-7 points), paragraph, executive (headline + bullets + recommendation). 30k char input. Mistral powered. | | text-classify | (0.005 USDC/call) Text classifier. Zero-shot classification — caller supplies 2-25 labels and the model picks the best one (or all matching labels in multi-label mode) with calibrated confidence scores and a per-label reason. Same backend as classify-text under a more search-friendly slug. Mistral powered. | | text-embedding | (0.002 USDC/call) Text embedding / vector embedding / semantic vector / Venice embeddings / Gemini embeddings / BGE-M3. Embeds 1 to 100 strings via Venice. Tier shorthand: 'default' → gemini-embedding-2-preview (newest, recommended), 'fast' → text-embedding-bge-m3, 'openai-compat' → text-embedding-3-small. You can also pass a full Venice embedding model name. Returns a list of vectors aligned with input order. | | text-normalize | (0.001 USDC/call) Text normalize. Unicode NFC / NFD / NFKC / NFKD normalization plus per-codepoint script classification (Latin / Cyrillic / Greek / Hebrew / Arabic / CJK / etc.), homoglyph detection (the Cyrillic 'а' that looks like Latin 'a'), and invisible / RTL / BOM detection. Same backend as unicode-normalize. Pure JS — no upstream API. | | thesaurus | (0.002 USDC/call) Thesaurus / synonyms / antonyms / similar words / rhymes / Datamuse / paraphrasing / query expansion. Returns related words for a given input word, configurable across five modes: 'synonyms' (rel_syn), 'antonyms' (rel_ant), 'sounds_like' (homophones / rhymes), 'similar_meaning' (semantically near), or 'related' (commonly co-occurring triggers). Wraps the public Datamuse API — no auth, commercial-OK. | | token-count | (0.001 USDC/call) Token count / tokenizer estimate / GPT-4 token count / Claude token count / Gemini token count / context-window pre-flight. Pure-local heuristic estimator targeting cl100k_base / o200k_base (GPT-4o) with calibrated multipliers for Claude and Gemini. Input model name is matched against an internal table of supported tokenizers. Accuracy ±5% versus tiktoken on typical English. No upstream call. | | tool-card-generate | (0.005 USDC/call) Tool card generator / OpenAI function-calling spec / A2A tool-card / agent tool description. Given a tool name + plain-English description (and optional parameter hints), emits a strict OpenAI/A2A-compatible tool card: { name, description, parameters: }. Powered by Venice zai-org-glm-4.7 in json_object mode. Useful for retrofitting LLM-callable tool descriptions onto existing API endpoints. | | translate | (0.005 USDC/call) AI translator. 100+ languages. Auto-detect source. ISO codes or names. Native fluency. Preserves Markdown / code / URLs. | | translate-text | (0.01 USDC/call) AI translator — 100+ languages, native-fluent. Auto-detects source. ISO codes or language names. Configurable formality (casual/formal/neutral). Preserves Markdown / code / URLs. | | type-inference-from-json | (0.001 USDC/call) Type inference from JSON / JSON to TypeScript / JSON to Zod / JSON to JSON Schema / JSON shape inferer / quicktype-style type generator. Paste a JSON sample (or array of samples); returns a generated type definition in your chosen format. Merges across array elements / object samples — properties present in some but not all become optional, mixed primitive types become unions. Auto-detects string formats (date-time / uuid / uri / email) for richer outputs. Pure compute, no upstream call, no API key. Ideal inside agent code-gen loops that need to consume an unfamiliar API response. | | unicode-normalize | (0.001 USDC/call) Unicode normalize / NFC NFD NFKC NFKD / homoglyph detection / IDN spoof / lookalike chars / invisible characters / zero-width / phishing detection / RTL override. Normalizes text to NFC (default), NFD, NFKC, or NFKD, classifies every codepoint by script (Latin / Cyrillic / Greek / Hebrew / Arabic / CJK / Hangul / etc.), flags Cyrillic / Greek / Latin Extended homoglyphs (the Cyrillic 'а' that looks like Latin 'a', etc.) with their position, codepoint, and the ASCII char they impersonate, and surfaces hidden / formatting characters like zero-width spaces, RTL overrides, and BOMs. Pure JS, no upstream API. |

How it works

1. Agent calls a tool (e.g. ai-to-human-text).
2. MCP server POSTs to https://x402.agentutility.ai/ai-to-human-text.
3. The endpoint responds HTTP 402 with payment instructions.
4. The MCP server signs an EIP-3009 USDC transfer authorization with X402_PRIVATE_KEY and retries.
5. CDP facilitator settles on Base.
6. The endpoint returns the actual response.

The agent never sees the payment flow — it just gets the result.

Links

• Cluster overview: https://agentutility.ai/wordmint/
• All MCP packages: https://mcp.agentutility.ai/
• Source: https://github.com/rooz21/x402/tree/main/packages/mcp-wordmint

Version: 0.2.2 · License: MIT