Live network intelligence
Observed across the agent economy.
The base rates behind agent infrastructure: what answers, how long listings have existed, which tools they expose, and where payment activity is visible. Everything below is counted from 2,105 research records and 710 unpaid probes in the observation archive as of 2026-08-11. Where the archive is thin, the chart says so.
Reachability: most of the catalogue does not answer.
Two independent measurements. Liveness comes from the observation sweep across every research record; the status taxonomy comes from unpaid HTTP probes against listed x402 endpoints.
Unpaid probe outcomes
69% of probes produced a qualifying response — 2xx, 3xx, or a 402 payment challenge. A 404 or 405 means the listed path is not where the catalogue says it is.
Counted from 710 unpaid probe records, latest 2026-08-12. A snapshot, not a trend.
Response latency of endpoints that answered
Mean 417ms. An agent paying per call feels this directly.
Measured response time recorded with each probe. A snapshot, not a trend.
Liveness by listing type
A registry entry is a claim, not a service. Across the observation sweep, 41% of listed MCP servers answered an initialize call, against 70% of x402 endpoint hosts.
Counted from the liveness sweep of 2026-08-11. A snapshot, not a trend.
Tenure: this market is new, and it shows.
Seller-owned domain registration age for x402 endpoint hosts, read from RDAP. Shared platform hosts are excluded from tenure credit, which is why the unknown column is large.
Domain registration age
40% of hosts with an attributable domain registered it in the last 90 days.
Domain registration dates read from RDAP for 1,556 endpoint hosts. A snapshot, not a trend.
Catalogue coverage
The archive tracks 36,484 catalogue entries. A research record — RDAP, wallet, llms.txt, tool inventory — exists for 5.8% of them so far.
Counted from catalogue history and the research records held today.
Tool surface: what an MCP server actually exposes.
Tool counts are inventoried by completing an initialize handshake and calling tools/list. A server that will not answer has no inventory, which is itself the finding.
Tools per MCP server
Inventoried for 247 of 549 servers with a research record.
Tool inventories recorded when each server was last researched.
Machine-readable posture of x402 hosts
Whether the seller has done the work an agent needs: documentation an LLM can read, and parseable payment terms.
Read from the research record held for each listing.
Money: almost nothing is moving yet.
Inbound USDC observed on Base at declared payee wallets, inside a disclosed block range. Observed inflow is not revenue, and a wallet declaration does not establish legal ownership.
Observed inflow by wallet
Every wallet with at least one observed inbound transfer, largest first. The entire observed market is 11.86 USDC.
Observed inbound USDC transfers on Base, scanned through 2026-08-11T16:34:15.565Z. Inflow is not revenue.
Wallet balance snapshots
A balance is not revenue and not inflow — it is a point-in-time reading at the declared address. Most declared wallets hold under one dollar.
Balance of each declared payment wallet at the last scan.
Observed modal price per call
The most common inbound transfer size at each active wallet — the closest observable proxy for a list price. Agent commerce is priced in fractions of a cent to a few cents.
Most common observed transfer size across 9 wallets. Observed transfers, not published prices.
Concentration: one wallet, many listings.
A declared payee wallet appearing across dozens of catalogue entries changes what a payment record proves. Model 1.0 discounts financial-evidence credit when the wallet is shared or the payer set is narrow.
Most-reused payee wallets
411 listings declare a payee across only 231 distinct wallets.
Payment addresses as declared by each listing, not proof of ownership.
Payer diversity at active wallets
How many distinct addresses actually paid. A single-payer wallet cannot distinguish a real customer from the operator testing their own endpoint.
Distinct paying addresses observed per wallet in the scanned block range.
Daily activity.
One point per archive day. The observatory archive opened 2026-08-11 and holds 2 day(s), so these series are as long as the record allows and no longer.
Catalogue entries added
New listings appearing in the tracked catalogues each archive day.
Counted from catalogue sweep history.
Probes run
Unpaid endpoint probes executed each archive day.
Counted from the unpaid probe log.
Observed USDC inflow
Inbound transfers to all scanned payee wallets each archive day.
Daily totals of observed inbound USDC transfers.
Why these are single points. The observatory began archiving 2026-08-11. A trend line drawn through one observation would be a decoration, not a measurement, so we draw the observation and label the gap. Each series extends automatically as archive days accumulate.
How to cite this page
MCP Scores / MCP Authority, observatory archive as of 2026-08-11, model 1.0. Underlying counts are reproducible through the public JSON API. If a figure here contradicts something you can measure, tell us — that is a correction, and we publish those.