For most of MCP's existence, servers have been free tools. An agent calls a tool, gets a result, nobody pays anything. That changed this batch. Roughly 25 new servers — more than 10% of the 232 new entries — involve money. Not financial data APIs. Payment infrastructure. The tools that let agents spend money, receive money, and verify that the money was well spent.
The Payment Layer
Three categories of payment infrastructure appeared simultaneously:
Escrow and settlement. agora402 and paycrow (both score 30) provide USDC escrow on Base — funds held in a smart contract until job completion, then released. agdel-mcp (score 35) implements a signal marketplace where agents create, buy, deliver, reveal, and settle information goods. These are not speculative concepts — they are deployed smart contracts with MCP interfaces.
Agent wallets and spending controls. clawvault (score 50) adds spending limits, whitelists, and human approval flows for agent wallets — the financial equivalent of sudo. Rill (score 57) takes a different approach: AI agent wallets with "Proof of Conduct" reputation and concentration decay on L1. The idea is that an agent's spending history becomes its credit score.
Service marketplaces. agoragentic (score 55) is an agent-to-agent marketplace — browse, invoke, and pay for AI services in USDC on Base. bstorms (score 44) sells agent playbooks — proven execution sequences that other agents can buy and follow. nexbid (score 39) provides product discovery via an open marketplace where advertisers bid for agent attention.
Pay-Per-Call Pricing
The harvey- suite from meltingpixelsai ships four specialized servers — budget management, market intel, web scraping tools, and post-transaction verification — all priced in USDC per API call. mifactory-bot takes the same approach with five servers: persistent agent memory ($0.01/write), web scraping ($0.005/page), document conversion ($0.10/convert), logic verification, and contract generation ($5-$20). These are SaaS businesses built as MCP servers with cryptocurrency payment rails.
x402 — the HTTP payment protocol that embeds USDC micropayments in standard HTTP headers — appears in several entries: botindex (17 x402-gated tools for sports betting and crypto intelligence, score 55), clawphunks (NFT minting with x402 payments, score 55), and near-solana-blockchain-api (24 blockchain endpoints with x402 USDC micropayments, score 22).
The Bitcoin Path
Not everything runs on Base. lightning-memory (score 57) combines decentralized agent memory with Lightning Network economics — Nostr identity for agents, L402 payments for memory access. sats4ai (score 22) offers Bitcoin-powered AI tools via Lightning micropayments with no signup or API keys required. These projects bet on Bitcoin's payment rails instead of Ethereum L2s.
What's Missing
Infrastructure. These servers provide the plumbing — escrow, wallets, marketplaces, pricing — but the autonomous agents that would use them barely exist yet. Today's AI agents run in supervised sessions where a human approves every significant action. An agent that autonomously browses a marketplace, evaluates service quality, negotiates price, pays in USDC, and verifies the result is still more vision than reality.
The builders are betting that the agents will catch up to the infrastructure. They may be right — the gap between what agents can do and what they're allowed to do is closing fast. But right now, the payment rails are ahead of the traffic.
Trust Scores Tell the Story
Most payment servers score in the 30-55 range — Low to Very Low Trust. This isn't because they're untrustworthy. It's because they're new, have few stars, and often lack the community signals that drive higher scores. The ones with higher scores (clawvault at 50, agoragentic at 55, rill at 57, lightning-memory at 57) tend to have better source code hygiene and more established repos. But none have the maturity signals that push servers above 60.
This is early infrastructure. Judge it by what it enables, not by its current polish.
Data sourced from the MCP Registry via MCP Scorecard. Trust scores computed from observable signals only.