io.github.veenrek/mapleai-mcp

MapleAI — GPT, Claude and Agent APIs with x402 Pay-Per-Request

OpenAI-compatible GPT and Anthropic-native Claude chat, multi-step agent execution, image, audio, X (Twitter) intel, embeddings and structured-decision APIs behind one endpoint. No accounts, no subscriptions, no API keys for the pay-per-request tier — an agent pays per call in USDC over the x402 protocol. Prepaid API keys are available for clients that prefer a token budget over per-request payments.

Live endpoints

Prepaid key endpoint (shared): https://mapleai.shop/v1

Models and pricing

GPT (per 1M tokens)

A request is quoted as counted input tokens × input rate + requested output tokens × output rate, plus the network settlement fee (about $0.001; the exact amount is always in the 402 challenge). Failed calls (HTTP ≥ 400) are cancelled, not settled.

Claude (per 1M tokens)

Native Anthropic Messages API at POST /v1/messages (system prompt, content blocks, SSE events, required max_tokens; the system prompt is billed as input tokens). Also reachable via the OpenAI-compatible /v1/chat/completions.

Agent execution

POST /v1/agents/execute — multi-step agent loop (reason → call tools → final answer), Cluster Protocol-compatible response. Optional SSE step streaming with "stream": true (step events as tools run, done with the final result). Built-in tools: deterministic calculator, data_analysis (statistics over up to 500 numbers), keyless web_search (Exa) and fetch_url. Per-step timeout returns partial results instead of a bare 502.

Charged as a ceiling — base + max_steps × per-step price (+ tool ceilings), like max_tokens for chat; up to 20 steps per task.

Images (per image)

POST /api/v1/images/generations and POST /api/v1/images/image2image (edit a PNG/JPEG/WebP data URI, max 10 MB). On Solana these routes settle upfront (image generation outlives a Solana blockhash) and retry transient upstream failures inside the same paid request.

Audio (per request)

TTS takes JSON {model, input, voice?, response_format?} and returns a WAV stream; OpenAI voice presets map to Gemini voices, the orpheus models use their own emotive voices (en: autumn/diana/hannah/austin/daniel/troy, ar: fahad/sultan/noura/lulwa/aisha/abdullah; [laughs]-style tags and real speed supported). STT takes multipart file + model or JSON base64, up to 25 MB; whisper-large-v3* add verbose_json word timestamps and srt/vtt. Responses carry x-audio-upstream and x-fallback-used headers — when the primary vendor fails the request fails over to a secondary one inside the same payment (for TTS the voice differs then).

X (Twitter) intel

Live X search and analytics via Grok x_search:

Jev structured decisions

POST /jev — jev-latest, $0.06 per 1M input tokens, output free. SystemOne protocol: send model, state and named questions (noul, choice or score with instructions), read answers from the response.

Free tier

POST /v1/embeddings — free NVIDIA nvidia/nemotron-3-embed-1b (2048 dimensions). Send input as a string or an array of up to 128 strings; model is optional and ignored.

POST /v1/free/chat/completions — free nvidia/gpt-oss-20b chat, rate-limited per agent (10 requests / 10 min and 100 / day).

Quickstart

# 1. Any request without payment returns HTTP 402 with the exact quote
curl -i https://base.mapleai.shop/v1/chat/completions \
  -H 'content-type: application/json' \
  -d '{"model":"openai/gpt-6-luna","messages":[{"role":"user","content":"Hello"}],"max_tokens":64}'

# 2. An x402-aware client signs the challenge and retries automatically

import { x402Client } from "@x402/core/client";
import { registerExactEvmScheme } from "@x402/evm/exact/client";
import { privateKeyToAccount } from "viem/accounts";

const client = new x402Client();
registerExactEvmScheme(client, { signer: privateKeyToAccount(process.env.EVM_PRIVATE_KEY) });

const first = await fetch(url, { method: "POST", headers, body });
const challenge = JSON.parse(Buffer.from(first.headers.get("payment-required"), "base64").toString());
const payment = await client.createPaymentPayload(challenge);
const paid = await fetch(url, {
  method: "POST",
  headers: { ...headers, "payment-signature": Buffer.from(JSON.stringify(payment)).toString("base64") },
  body,
});

A runnable end-to-end demo (free embeddings → paid GPT answer) lives in examples/embeddings-to-chat/ and online at https://base.mapleai.shop/examples/embeddings-to-chat/.

Building an agent? Fork the agent-starter/ template — repo instructions for agent clients (CLAUDE.md/AGENTS.md), MCP wiring, a full paid scenario script (free embeddings → prepaid tap → prepaid chat), and a free CI smoke for all four gateways.

Prepaid API keys

Buy a prepaid bearer key for one GPT or Claude model, or a Jev input pack, with a single x402 payment:

curl https://base.mapleai.shop/prepaid/codes \
  -H 'content-type: application/json' \
  -d '{"model":"openai/gpt-6-luna","tokens":100000}'

Budgets run in 100 000-token steps from 100 000 to 1 000 000, priced at the model input rate plus the settlement fee:

GPT and Claude keys spend at POST /prepaid/v1/chat/completions (works on the subdomains and at the shared https://mapleai.shop/v1, OpenAI-compatible); Jev packs spend at POST /prepaid/v1/jev. Check usage and status for free:

curl https://mapleai.shop/v1/prepaid/status \
  -H "Authorization: Bearer oms_buy_..."

Discovery

• GET /v1/models — free model catalog with live pricing
• GET /.well-known/x402 — x402 resource manifest
• GET /openapi.json — OpenAPI 3.1 specification
• GET /llms.txt, GET /AI-AGENTS.md — agent-oriented docs
• GET /developers — developer guide with examples

Repository layout

• src/ — the x402 gateway (Express, TypeScript, runs under tsx)
• admin/ — OmniRoute-based admin app: provider connections, combos, prepaid key issuance and wallet-only (SIWE) dashboard
• examples/ — runnable client examples
• packages/mapleai-mcp/ — local stdio MCP server (free model catalog, embeddings and prepaid status; paid GPT and Claude chat, Jev, agent execution and prepaid key packs with local x402 signing; bought keys spend in-server via prepaid_chat with no wallet)
• tools/ — payment and upstream test scripts
• deploy/ — systemd units and Apache vhosts used on the VDS

Development

npm install
cp .env.example .env   # fill in UPSTREAM_API_KEY, PAY_TO, model prices
npm run typecheck
npm start              # listens on PORT (default 4021)

Required env: UPSTREAM_API_KEY, PAY_TO, MODEL_PRICES, MODEL_MAPPING. Optional features are enabled by their own keys (IMAGE_UPSTREAM_API_KEY, NVIDIA_API_KEY, JEV_UPSTREAM_API_KEY, PREPAID_ISSUER_TOKEN, …) — see .env.example for the full list.