Paid MCP Tools: 65 Listings, 12.5 Calls a Day, and One Site Making Seven Times the Whole Market

A dedicated x402 transport specification defines how an MCP tool charges for a call, and it is complete. Two full enumerations of the discovery index, two days apart, show what it is being used for: almost nothing, and slightly less than before.

The whole paid-MCP market, counted twice

| | 3 Aug 2026 | 5 Aug 2026 | |---|---|---| | Listings mentioning MCP | 68 | 65 | | Share of the index | 0.46% | 0.44% | | Calls, trailing 30 days | 389 | 374 | | Calls per day, whole market | 13.0 | 12.5 | | Distinct payers, trailing 30 days | 113 | 111 |

Every one of the 65 has at least one call, so this is not a graveyard of dead listings — it is a live market that is simply tiny. For scale, Sinapsi's own MCP endpoint served 84.6 requests a day over the trailing seven days: one site, unpaid, handling 6.8× the traffic of every paid MCP tool in the index combined.

Where the calls are

| Calls (30d) | Host | |---|---| | 125 | aeml-x402.zeabur.app | | 89 | mcp.deep3.ai | | 60 | mcp.yield.xyz | | 23 | mcp.barker.money | | 11 | x402-mcp.onrender.com |

Two of the top five are hosted on free or hobby tiers of deployment platforms (zeabur.app, onrender.com). This matches what we found across the whole index, where 21 listings were served from a free ngrok tunnel and had vanished by the next snapshot — our own count, by substring match on the resource URL across the two enumerations declared above (The x402 Market, Counted: 14,766 Listings, 12,741 Calls a Day, One Winner): a production discovery index carrying a large amount of somebody's evening project.

Paid MCP is priced ten times cheaper than the rest of the market

Across 184 declared MCP offers the median price is $0.0010 — a tenth of a cent — against a $0.010 median for the index as a whole. The range runs from $0.0010 to $2.00, but the mass sits at the floor.

This is the interesting anomaly. MCP tools are, in principle, the *more* valuable shape: a tool call is an action inside an agent's reasoning loop, chosen deliberately, whereas an HTTP fetch is a document retrieval. Sellers are pricing the deliberate act at a tenth of the price of the document. That is not what you would expect if anyone believed the tool call was worth more.

The mechanism is not the obstacle

The binding does not live in the main v2 specification — it is a separate transport document, specs/transports-v2/mcp.md, which we retrieved and read in full on 5 August 2026 (x402: HTTP 402 Finally Gets a Job, at 32 Cents a Transaction). Quoted verbatim:

- the 402 is a tool result, not a transport error: *"When a tool requires payment, servers MUST return a tool result with isError: true containing the PaymentRequired data"*, whose mechanism is *"Tool result with isError: true, structuredContent, and content fields"*. So the model sees a priced offer rather than a broken connection; - both carriers are mandatory — *"structuredContent (REQUIRED): Direct PaymentRequired object"* and *"content[0].text (REQUIRED): JSON-encoded string of the same PaymentRequired object"* — the second for clients that cannot read structured content; - the signed payment goes up in the request: *"Clients send payment data using the MCP _meta field with key x402/payment"*; - the receipt comes back down: *"Servers communicate payment settlement results using the _meta["x402/payment-response"] field"*.

Nothing here needs inventing. A seller can charge for an MCP tool today.

Our reading of why it is empty

This section is interpretation, not measurement.

An MCP server is typically *installed* by a human, once, with credentials the human already holds. The agent then calls it as often as it likes at no marginal cost. The whole shape of MCP adoption to date is subscription-like and human-mediated, and a per-call charge cuts against it — the buyer at the moment of purchase is a human choosing a server, not an agent choosing a call.

x402 assumes the opposite: an agent with a wallet, discovering an unknown seller at the moment of need. We can put a number on how rare that buyer is, rather than assert it: across the whole paid-MCP subset in thirty days there were 111 distinct payers, against 374 calls (The Earn-Spend Loop: Why Machine Payment Is Half-Built). Whatever the reason, the discovering-and-paying agent is at present a population of roughly a hundred wallets worldwide.

Both readings point the same way: the empty channel is cheap to enter and, for now, worth little. Being early here costs almost nothing and proves almost nothing. It is worth doing because the entry cost is near zero, not because the traffic is there — and anyone claiming otherwise is selling something (What Is Actually Priced: The One-Cent Ceiling, and the Market That Escaped It, The Agent Labour Market, Probed: A Complete Bounty Board With Four Customers).

Falsifier

We are wrong to call this empty if, by 30 June 2027, MCP listings in the discovery index exceed 2% of the index or the subset's calls exceed 500 per day. Both are recomputable from a fresh snapshot with the script in agent-economy/dati/.

We are also wrong if the substring filter is hiding the market: if a census of MCP servers built by some other method finds ten times more paid MCP endpoints than the URL match does, this page is measuring our filter rather than the market. That check we have not done, and it is the weakest link in the page.