What are agents buying in Coinbase’s x402 Bazaar discovery layer?

The agentic economy is actively running and working, but what exactly are these agents buying?

The short answer is data. Agents buy information: web search, market and token data, contact records, social posts and chain state. Almost nothing in the catalog is a finished good or service. Instead, it’s machine-readable data priced in fractions of a cent.

How agents use discovery layers 

Just like a person browses an app store, an agent queries a discovery layer in code: it asks for a capability, gets back a list of endpoints with prices and usage stats, and calls the one it wants.

Paying happens through the x402 protocol. The agent calls the service and receives an HTTP 402 Payment Required response with a price and payment details. It pays the amount in USDC from its own wallet, a facilitator settles the payment onchain, and the agent retries the same call with proof of payment to get the result. The whole loop, discover, pay, receive, runs with no person in it.

Agent purchases by the numbers

Between August 13 and September, 13, 2026, x402 on Base has recorded 44,930,157 authorized stablecoin transfers from 440,608 distinct payers, at a median payment of $0.0061. 

Here is how the top 100 x402 Bazaar listings break down by 30-day call volume:

The sections below organize some of these listings by category, describing what each service is and how an agent uses it. 

Web search and content retrieval

This is the single largest category by call volume, roughly 88,000 calls across the top 100. The single largest share of these calls goes to two upstream providers, Exa (neural web search) and Firecrawl (crawl and scrape).

Some of the top services include:

How agents use it:

An agent that needs current information it was not trained on calls one of these to search the live web, then reads the top pages as clean markdown before reasoning or acting. It is the closest thing to a default first step in an agent loop, which is why it dominates.

Crypto market data, token analytics, and trading signals

As the second-largest cluster in x402 payments, these services feed trading and research agents with token prices, safety scores, smart-money flows, liquidation maps, and swap quotes. Several of the top services are Solana focused.

However, it’s important to note the payer counts: most of these feeds serve a handful of addresses, so this category is a few trading operators running high-frequency loops, not a broad market.

How agents use it:

A trading agent screens a token for safety and concentration, checks smart-money and whale flows, prices a swap, and decides whether to act, all without a human. 

People and company enrichment

The B2B sales-data category. Agents resolve a name, email, or LinkedIn URL into contact details, company data, and dossiers. Almost everything here wraps a known data vendor (People Data Labs, FullEnrich, Clado, Apollo, Whitepages).

How agents use it:

This is the engine behind autonomous prospecting. A sales agent takes a list of names or companies, enriches each into a verified contact, and hands back a lead list or populated CRM records. These are also the highest-priced calls in the catalog (the PDL match bills $0.28, Whitepages $0.22, Clado $0.20), because the underlying data is expensive.

Social media search

Agents use these services to query Twitter/X and Reddit for real-time posts, mostly for monitoring, alerting, and sentiment.

How agents use it:

Monitoring agents watch a handle, hashtag, or topic and trigger alerts. Research agents pull recent posts to gauge sentiment before a trade or a report. This is a lightweight, high-frequency read pattern.

Onchain RPC and blockchain reads

By far the widest genuine adoption in the catalog. Each onesource.io RPC endpoint serves hundreds of distinct payers, and its ERC-20 balance endpoint reaches 1,070 distinct payers. Most search and trading feeds serve far fewer payers, from a handful to a couple hundred. Not every read here is broad, though: api.zerion.io saw 451 calls from just two payers.

How agents use it:

Agents that need to read chain state without running a node call these to check a balance, resolve an ENS name, confirm a transaction, or inspect a contract before acting. Across the onesource endpoints the ratio of calls to payers is close to one to one, demonstrating that many different agents make just a few calls each, which is what real, distributed usage looks like.

AI inference

Agents are also accessing sub-agents, specialized AI workers that a primary agent calls to handle specific tasks and pay per call rather than maintain a subscription. Of note, one listing runs inference inside an attested enclave, keeping the prompt confidential even from the operator.

How agents use it:
An orchestrating agent offloads a reasoning or drafting step to a model it pays for per call, with no account or key to manage. The enclave option matters for agents handling sensitive prompts, since only the enclave can decrypt the message.

Niche and long-tail services

Below the top categories, the catalog spreads into single-purpose services that see real but small use. Notable examples:

• Travel: stabletravel.dev searches cached Seats.aero award-flight availability (4,122 and 1,103 calls), so a travel agent can find business-class award seats by route and date.
• Onchain loyalty: api.loyalspark.online exposes a full rewards platform (balances, vouchers, transfers) that agents automate on behalf of a business, across roughly six endpoints in the 400 to 1,100 call range.
• SEO and keyword data: google-trends.use.x402atlas.com returns Google Trends interest and related queries (1,250 and 493 calls) for content and demand research.
• News: x402.ottoai.services and x402stock.xyz serve ranked crypto and equity headlines with timestamps and sentiment for market-monitoring agents.
• Specialized data: api.geoprimitives.dev answers nautical-chart depth queries from NOAA ENC data, and stockwaves.net runs a pre-trade authenticity check on tokenized US stocks.

Turnkey: Securing agentic transactions, large and small

The x402 Bazaar shows what one form of agentic commerce looks like in practice: frequent, low-cost micropayments for specific pieces of information or compute that an agent needs to complete a task.

That autonomy changes the security model. An agent with a funded wallet can make purchases without a person approving each one, so developers need enforceable limits on what it can buy, how much it can spend, where it can send funds, and how often it can transact.

As discovery layers such as the x402 Bazaar make services easier for agents to find and purchase, programmable wallet controls become essential infrastructure. The goal is to let agents transact autonomously within boundaries that developers and users can define, enforce, and audit.

Explore Turnkey’s wallet infrastructure for AI agents and get started with Turnkey today.