mcpField
B
Blixtworks

Blixtworks

Remote
blixtworks.comAI & Agents0Updated 12w ago

About Blixtworks

Blixtworks provides AI agents with a pay-per-call vision toolbox and a marketplace worker. It accepts USDC on Base through x402 and includes image categorization, captioning, OCR, and embeddings.

Key features

  1. 01Zero-shot image labels and confidence
  2. 02Generate image captions
  3. 03Extract printed English text
  4. 04Generate 512-dimensional CLIP vectors
  5. 05Automatic x402 payments for MCP tools

Use cases

  1. 01Organize images by category
  2. 02Create short descriptions for images
  3. 03Extract printed English text from images
  4. 04Call vision tools and pay per use through MCP

MCP support

The MCP server offers 10 tools and handles x402 payments automatically for MCP clients. The README also lists four vision endpoints: categorization, captioning, OCR, and embeddings.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://www.blixtworks.com/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Packages
npmblixtworks-mcp1.1.0

Tools

68
ParameterTypeRequiredDescription
groupstringNovision, media, text, data, web, convert or chain

Details

From the project README

img-categorize / agent-hustle

An autonomous money system: a pay-per-call vision toolbox for AI agents, plus a marketplace worker. Payments via x402 — USDC on Base, no accounts, no API keys, settled straight to a self-custody wallet.

Live: https://www.blixtworks.com

The toolbox

EndpointDoesPrice
POST /categorizezero-shot labels + confidence, custom label sets$0.02
POST /captionone-sentence image description$0.02
POST /ocrprinted-text extraction (English)$0.03
POST /embed512-dim CLIP vector$0.01

Free: GET /demo, GET /openapi.json, GET /.well-known/x402, landing page. Buyers are never charged for failed requests (settlement cancels on any 4xx/5xx).

Architecture

  • Front door (Render free tier, INFERENCE=proxy): stable URL, x402 paywall, ~100MB RAM, no models. Auto-deploys from this repo's master.
  • Inference box (any machine with RAM, INFERENCE=local): runs the actual models (CLIP, ViT-GPT2, Tesseract) behind /infer, exposed via a cloudflared quick tunnel.
  • The worker announces the current tunnel URL to the front door every few minutes via /internal/upstream, signed with the agent wallet key and verified on-chain-style (viem verifyMessage) — no shared-secret setup, no hijackable upstream.
  • Front door offline-safe: no upstream → 503 → buyer not charged.

Run the home side

./start.sh          # server (local mode) + tunnel + worker, idempotent
node src/status.js  # balances, sales, listings, current URLs

The worker (every 10 min): wallet balances, ClawTasks bounty scan, Bazaar market intel, 402index listing upkeep, signed upstream heartbeat (plus a fast 4-min heartbeat timer).

Distribution

MCP server: mcp/ — 10 tools with automatic x402 payment for any MCP client.

Listed on x402scan (SIWX wallet-signature registration: node src/register-x402scan.js <url>) and 402index (keyless, worker maintains it). PR to awesome-x402 pending. Moltbook agent ImgCatWorker registered (first post queued pending human claim).

Wallet

Self-custody, generated locally with viem — no exchange, no KYC. Address in .env (AGENT_ADDRESS), private key in .env (AGENT_PRIVATE_KEY, mode 600). Back up .env — it is the money. Import into any wallet app (network: Base) to withdraw.

Ground rules

No botting human-only platforms, no fake engagement, no spam. This system sells honest compute and works agent-native marketplaces that explicitly welcome AI workers.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.