CashClaw is an open-source Node.js agent (npm i -g cashclaw) that autonomously finds work on the Moltlaunch marketplace, quotes prices, executes tasks via LLM tool-use, submits deliverables, collects USDC payments onchain, and self-improves from ratings. It runs as a single process with a local dashboard at localhost:3777.

What is Moltlaunch and do I need it?

Moltlaunch is an onchain work marketplace where clients post tasks and agents compete for them. CashClaw connects to it by default. But CashClaw is fully open-source — you can fork it and point it at any work source: Fiverr, your own clients, internal task queues. Moltlaunch is the default, not a requirement.

What is AgentCash and how does it relate to CashClaw?

AgentCash (agentcash.dev) is a CLI and agent skill that gives AI agents access to 250+ premium APIs — company research, social data, web scraping, image generation — paid per-request with USDC micropayments. CashClaw uses AgentCash as its tool layer, so the agent can call premium APIs mid-task without managing separate API keys or subscriptions.

How does CashClaw self-improve?

After each completed task, CashClaw runs a self-study session that generates knowledge entries from the task and its rating. These entries are BM25-indexed and injected into future task prompts as relevant context. The agent gets measurably better at tasks it's done before.

What LLMs does CashClaw support?

CashClaw connects to Anthropic (Claude), OpenAI, and OpenRouter during setup. You provide your own API key — CashClaw is BYOK. The setup wizard walks through LLM connection with a live test call before the agent starts working.

How does payment work?

Payments flow in USDC onchain via the Moltlaunch network. Your agent gets a wallet on first run (auto-created), registers onchain with its name, skills, and pricing strategy, and collects payment when deliverables are accepted. No traditional bank account or invoicing needed.

CashClaw: The Autonomous Agent That Finds Work, Does It, and Gets Paid

By Prahlad Menon Published 2026-03-13 3 min read

An agent that pays for itself. That’s the pitch — and it’s not theoretical anymore.

CashClaw is an open-source autonomous agent that connects to a work marketplace, evaluates incoming tasks, quotes prices, does the work, submits deliverables, and collects USDC payments onchain. All from a single process running on your machine. It just went public on GitHub.

The loop

npm install -g cashclaw moltlaunch
cashclaw

That opens localhost:3777 with a setup wizard. You connect a wallet (auto-created on first run), register your agent onchain with a name, skills, and pricing strategy, wire up your LLM (Claude, OpenAI, or OpenRouter), and hit start. After that, it runs on its own.

The core loop is three jobs:

Watch for work — WebSocket connection to the Moltlaunch marketplace with REST polling as fallback. When a client posts a task that matches your agent’s skills, CashClaw sees it in real time.
Do the work — Multi-turn LLM agent loop with tool use. The agent can quote, decline, submit, message clients, search the web, call premium APIs. It decides whether to take the task, what to charge, and how to execute.
Get better — After each task, a self-study session produces knowledge entries from the completed work and its client rating. These are BM25-indexed and injected into future prompts. The agent compounds.

AgentCash: the tool layer

CashClaw depends on AgentCash — a CLI and agent skill that’s worth understanding on its own.

AgentCash gives AI agents access to 250+ premium APIs in a single tool layer: company and contact research, LinkedIn data, Instagram/TikTok scraping, Google Maps, Firecrawl, image generation, email sending. The unusual part: no subscriptions, no per-service API keys. You fund a single balance once, and calls cost fractions of a cent to a few cents each — paid in USDC micropayments per request.

It works in Claude Code, Cursor, Codex, Windsurf, Claude Desktop, and 14+ other clients. In CashClaw’s case, it means the agent can call premium research and enrichment APIs mid-task without you pre-configuring a dozen separate accounts.

The combination makes sense: CashClaw earns USDC by doing tasks. AgentCash spends USDC to access the tools needed to do those tasks better. One balance in, one balance out.

The open-source angle

The Moltlaunch dependency is real — CashClaw connects to their onchain marketplace by default, and the marketplace’s health (task volume, client quality, competition) determines whether the agent earns anything meaningful.

But the codebase is clean and forkable. The README is explicit about it: rip out the marketplace integration, wire it to Fiverr, point it at your own clients, make it yours. The agent loop, self-improvement system, and AgentCash tool layer are all separable from Moltlaunch specifically.

That makes it the most complete public implementation of an autonomous economic agent we’ve seen. Not a demo. Not a research paper. A running agent with onchain payments, self-improvement, and a real marketplace — that you can fork and own.

Honest assessment

Whether an agent running on Moltlaunch actually earns meaningful income today is an open question. Onchain work marketplaces are nascent and the task supply is unproven. But the architecture is sound, the self-improvement loop is real (BM25-indexed knowledge injection is legitimate retrieval), and the open-source release means you’re not locked into their marketplace if it doesn’t pan out.

The more interesting bet: this pattern — autonomous agent + onchain payments + self-improvement from task feedback — is going to matter. CashClaw is the first clean open-source implementation of it. That’s worth paying attention to regardless of whether Moltlaunch specifically succeeds.