Blockscout published an argument that the new API consumer is an agent, and it is unusually concrete about what that costs an operator. “Agents don’t behave like standard browser traffic. They behave like a load test that never ends.” The diagnosis underneath is about context rather than volume: “An agent re-reads the tool schema on every session, has no persistent memory of your API’s quirks, and burns its own context window parsing whatever you hand back.” Their answer ships as infrastructure rather than advice — an MCP server at mcp.blockscout.com/mcp exposing 16 tools including get_address_info, get_transaction_info and read_contract, live across 120+ chains, with two agent skills published alongside it, and x402 payments live on Base using EIP-3009 and a plain HTTP 402 to charge per call rather than per key.
Take the framing seriously and the metered part is the sharper half. A 402 on the request path is a pricing model that does not require an agent to hold an account, which is the practical blocker for autonomous use of any paid API. It is also worth being precise about scope: x402 is live on Base specifically, not everywhere the explorer runs, and the Pro API key remains the conventional path.
Here the catalog has to report on itself, because it disagrees with the announcement. Blockscout carries 18 API pages — the REST API v2, the GraphQL API, the Etherscan-Compatible API, the PRO API and the resource-level surfaces beneath them — and its agentic access profile counts 57 operations with 1 acting, a near-pure read estate, which is exactly right for a block explorer. What the record does not contain anywhere is the MCP server. No MCP artifact, no agent skills, no x402 pointer, nothing in the registered contract that mentions any of it.
Blockscout scores 27.4, thin on the Kin Score, on discoverability at 75.9 and contract quality at 45.5 against governance at 0.0. Agent Readiness is 22.3, agent-aware, and mcp_server and agent_skills both read unlit — for a provider that runs a public MCP server with 16 tools and publishes two agent skills. That reading is accurate against what Blockscout has registered and wrong about what Blockscout has built, and the gap between those two sentences is the whole job of a catalog. Three of the four unlit dimensions closest to this story are closed by a pointer, not by engineering. We will go re-read their surface; the more useful outcome would be Blockscout declaring it, because then every catalog finds it, not just the one that read the blog post.