<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom"><title>MCP and x402 measurements — modelmarket.dev blog</title><link href="https://modelmarket.dev/blog/"/><link rel="self" href="https://modelmarket.dev/blog/feed.xml"/><id>https://modelmarket.dev/blog/</id><updated>2026-11-13T21:21:00Z</updated><author><name>Alex</name></author><entry><title>Selling eSIMs to AI agents: the seller&#x27;s side, tried on a laptop in 32 seconds</title><link href="https://modelmarket.dev/blog/selling-esims-to-ai-agents-tried-on-a-laptop-in-32-seconds/"/><id>https://modelmarket.dev/blog/selling-esims-to-ai-agents-tried-on-a-laptop-in-32-seconds/</id><published>2026-11-13T21:21:00Z</published><updated>2026-11-13T21:21:00Z</updated><summary>For eSIM resellers: ./onboard.sh --try runs the storefront and the agent-facing seller node on localhost with a sample catalogue, no supplier account, domain or keys. What an agent sends, what a selling node does with the payment, and what is still untested.</summary></entry><entry><title>Six of 16 common MCP servers can&#x27;t be read from outside. One of the others changed 19 times in a month</title><link href="https://modelmarket.dev/blog/six-of-16-common-mcp-servers-cant-be-read-from-outside/"/><id>https://modelmarket.dev/blog/six-of-16-common-mcp-servers-cant-be-read-from-outside/</id><published>2026-11-12T05:57:00Z</published><updated>2026-11-12T05:57:00Z</updated><summary>30 days of tool-definition history for 16 MCP servers we picked as a sample team stack: what changed, what the version missed, what stays hidden.</summary></entry><entry><title>Same MCP version. Different tool description. Who signed it?</title><link href="https://modelmarket.dev/blog/same-mcp-version-different-tool-description-who-signed-it/"/><id>https://modelmarket.dev/blog/same-mcp-version-different-tool-description-who-signed-it/</id><published>2026-11-11T07:27:51Z</published><updated>2026-11-11T07:27:51Z</updated><summary>A one-line tools/list edit changes the manifest without a version bump. Follow that change through SHA-256, Sigstore, Rekor, and HISTOR.</summary></entry><entry><title>A missing version broke verification of signed ARD search results</title><link href="https://modelmarket.dev/blog/a-missing-version-broke-verification-of-signed-ard-search-results/"/><id>https://modelmarket.dev/blog/a-missing-version-broke-verification-of-signed-ard-search-results/</id><published>2026-11-07T20:42:47Z</published><updated>2026-11-08T07:54:56Z</updated><summary>Trace discovery to invocation, then reproduce the missing-version failure while keeping the JWS and descriptor digest valid.</summary></entry><entry><title>A privacy warning became bad advice in an AI reviewer&#x27;s draft</title><link href="https://modelmarket.dev/blog/a-privacy-warning-became-bad-advice-in-an-ai-reviewers-draft/"/><id>https://modelmarket.dev/blog/a-privacy-warning-became-bad-advice-in-an-ai-reviewers-draft/</id><published>2026-11-07T20:42:47Z</published><updated>2026-11-08T07:54:56Z</updated><summary>A lost advisory flag turned a privacy note into bad advice. The adapter diff, filtering contract and two recorded examples show the fix.</summary></entry><entry><title>Healthy containers hid a blocked MCP marketplace connection</title><link href="https://modelmarket.dev/blog/healthy-containers-hid-a-blocked-mcp-marketplace-connection/"/><id>https://modelmarket.dev/blog/healthy-containers-hid-a-blocked-mcp-marketplace-connection/</id><published>2026-11-07T20:42:47Z</published><updated>2026-11-08T07:54:56Z</updated><summary>Pinned definitions, live prices and a healthy process: a five-case check and the MCP admission states behind a silent outage.</summary></entry><entry><title>An MCP tool set can gain a digest without changing</title><link href="https://modelmarket.dev/blog/an-mcp-tool-set-can-gain-a-digest-without-changing/"/><id>https://modelmarket.dev/blog/an-mcp-tool-set-can-gain-a-digest-without-changing/</id><published>2026-11-07T16:00:00Z</published><updated>2026-11-08T07:54:56Z</updated><summary>Eight fixtures and a runnable Python example show how MTL handles numeric schemas, missing digests and profile revisions.</summary></entry><entry><title>242,072 malware flags on npm and PyPI, and what the registries did about them</title><link href="https://modelmarket.dev/blog/242072-malware-flags-on-npm-and-pypi/"/><id>https://modelmarket.dev/blog/242072-malware-flags-on-npm-and-pypi/</id><published>2026-11-03T13:57:00Z</published><updated>2026-11-03T13:57:00Z</updated><summary>We measured what open package-malware feeds flag on npm and PyPI, what the registries did, how many flags left up are wrong, and what open scanners cost in false alarms.</summary></entry><entry><title>We found 331 MCP tool-list changes in one week that only moved a date. Each one costs a prompt-cache write</title><link href="https://modelmarket.dev/blog/331-mcp-tool-list-changes-only-moved-a-date/"/><id>https://modelmarket.dev/blog/331-mcp-tool-list-changes-only-moved-a-date/</id><published>2026-11-02T09:10:04Z</published><updated>2026-11-02T09:10:04Z</updated><summary>Of 10,712 tool-definition changes HISTOR recorded in a week, 331 changed nothing but a date or a timestamp. Where those dates come from, and where they belong instead.</summary></entry><entry><title>We read the robots.txt of 2,347 MCP hosts: Python 3.12&#x27;s parser gets 252 endpoints wrong</title><link href="https://modelmarket.dev/blog/we-read-the-robots-txt-of-2347-mcp-hosts/"/><id>https://modelmarket.dev/blog/we-read-the-robots-txt-of-2347-mcp-hosts/</id><published>2026-11-02T09:10:04Z</published><updated>2026-11-02T09:10:04Z</updated><summary>urllib.robotparser took the first matching rule and read * and $ literally until 2026. Python 3.14.5 and 3.13.14 fixed it; 3.12 never will. On MCP endpoints the difference is 14%.</summary></entry><entry><title>We stopped reading 2,195 MCP servers for three days. Another team&#x27;s log is how we found out</title><link href="https://modelmarket.dev/blog/we-stopped-reading-2195-mcp-servers-for-three-days-another-teams-log-is-how-we-found-out/"/><id>https://modelmarket.dev/blog/we-stopped-reading-2195-mcp-servers-for-three-days-another-teams-log-is-how-we-found-out/</id><published>2026-11-02T04:24:01Z</published><updated>2026-11-02T04:24:01Z</updated><summary>Our MCP crawler tried the 2026-07-28 protocol first and refused to fall back. Per-read success stayed at 99%. A second, independent log showed what we were missing.</summary></entry><entry><title>An honest x402 seller earned $0.18 in two weeks and shut down. We now host sellers like it for free.</title><link href="https://modelmarket.dev/blog/an-honest-x402-seller-earned-018-in-two-weeks-and-shut-down-we-now-host-sellers-like-it-for-free/"/><id>https://modelmarket.dev/blog/an-honest-x402-seller-earned-018-in-two-weeks-and-shut-down-we-now-host-sellers-like-it-for-free/</id><published>2026-11-01T06:21:29Z</published><updated>2026-11-01T06:21:29Z</updated><summary>It answered correctly, charged less than we do, earned $0.18 in two weeks and went offline. Why small x402 sellers fold, and the free hosting we now offer them.</summary></entry><entry><title>Two public records of MCP tool definitions, one day, 23,190 servers: where they agree</title><link href="https://modelmarket.dev/blog/two-public-records-of-mcp-tool-definitions-one-day-23190-servers-where-they-agree/"/><id>https://modelmarket.dev/blog/two-public-records-of-mcp-tool-definitions-one-day-23190-servers-where-they-agree/</id><published>2026-10-31T17:56:44Z</published><updated>2026-10-31T17:56:44Z</updated><summary>HISTOR and heldfast read the same MCP servers a minute apart: 99.1% of 318,389 tools match byte for byte. The rest shows why a digest needs its capture conditions.</summary></entry><entry><title>We built a marketplace for AI agents. Sellers told us the integration wasn&#x27;t worth it.</title><link href="https://modelmarket.dev/blog/we-built-a-marketplace-for-ai-agents-sellers-told-us-the-integration-wasnt-worth-it/"/><id>https://modelmarket.dev/blog/we-built-a-marketplace-for-ai-agents-sellers-told-us-the-integration-wasnt-worth-it/</id><published>2026-10-31T09:57:09Z</published><updated>2026-10-31T10:06:42Z</updated><summary>Two API sellers declined our federation proposal. Their reasons exposed a cost our protocol didn&#x27;t account for—and changed what we built next.</summary></entry><entry><title>We Improved WARDEN&#x27;s Warnings. Twelve Rechecks Changed No Choices.</title><link href="https://modelmarket.dev/blog/we-improved-wardens-warnings-twelve-rechecks-changed-no-choices/"/><id>https://modelmarket.dev/blog/we-improved-wardens-warnings-twelve-rechecks-changed-no-choices/</id><published>2026-10-25T19:09:17.681814+00:00</published><updated>2026-10-25T19:09:17.681814+00:00</updated><summary>720 fresh model answers showed a version-level improvement on six known cases. Twelve Stop rechecks changed no selected IDs. Here is what each result means.</summary></entry><entry><title>A copyable MCP tool-cost check for CI, with passing and failing examples</title><link href="https://modelmarket.dev/blog/a-copyable-mcp-tool-cost-check-for-ci-with-passing-and-failing-examples/"/><id>https://modelmarket.dev/blog/a-copyable-mcp-tool-cost-check-for-ci-with-passing-and-failing-examples/</id><published>2026-10-25T08:02:01Z</published><updated>2026-10-25T09:26:47Z</updated><summary>Pinned commands, the complete snapshot fetcher, and a separate gate for total tokens, descriptions and input schemas.</summary></entry><entry><title>Five Wallets Paid an x402 Seller. What Can We Say About Demand?</title><link href="https://modelmarket.dev/blog/five-wallets-paid-an-x402-seller-what-can-we-say-about-demand/"/><id>https://modelmarket.dev/blog/five-wallets-paid-an-x402-seller-what-can-we-say-about-demand/</id><published>2026-10-25T07:47:20Z</published><updated>2026-10-25T09:26:44Z</updated><summary>A five-wallet audit verifies 24 settlements while separating payment evidence, behavioral signals and unknown customer intent.</summary></entry><entry><title>We Kept WARDEN&#x27;s 8/10 Thresholds and Fixed Its Credential Index</title><link href="https://modelmarket.dev/blog/we-kept-wardens-810-thresholds-and-fixed-its-credential-index/"/><id>https://modelmarket.dev/blog/we-kept-wardens-810-thresholds-and-fixed-its-credential-index/</id><published>2026-10-25T07:47:17Z</published><updated>2026-10-25T07:47:17Z</updated><summary>A reader asked whether shorter secret fragments justified lower thresholds. We compared 6/7/8, reproduced false blocks, and fixed a narrower indexing defect.</summary></entry><entry><title>We rechecked 292 failing MCP packages. 12 newer releases now list tools.</title><link href="https://modelmarket.dev/blog/we-rechecked-292-failing-mcp-packages-12-newer-releases-now-list-tools/"/><id>https://modelmarket.dev/blog/we-rechecked-292-failing-mcp-packages-12-newer-releases-now-list-tools/</id><published>2026-10-25T07:43:53Z</published><updated>2026-10-25T09:26:41Z</updated><summary>A fresh-install follow-up separates release recovery, optional dependencies and a correction to our own SDK classifier.</summary></entry><entry><title>We attacked our own MCP firewall with its source code open. 14 of 17 attacks got through; now 2 of 23 do</title><link href="https://modelmarket.dev/blog/we-attacked-our-own-mcp-firewall-with-its-source-code-open-14-of-17-attacks-got-through-now-2-of-23-do/"/><id>https://modelmarket.dev/blog/we-attacked-our-own-mcp-firewall-with-its-source-code-open-14-of-17-attacks-got-through-now-2-of-23-do/</id><published>2026-10-23T09:40:00Z</published><updated>2026-10-23T09:40:00Z</updated><summary>Asked what WARDEN still misses, we wrote 17 attacks on its MCP flow guard with the source open. 14 got through; after the fixes, 2 of 23 do. The rest is named.</summary></entry><entry><title>MCP servers edit the instructions your agent reads. WARDEN 0.18 shows which words changed</title><link href="https://modelmarket.dev/blog/mcp-servers-edit-the-instructions-your-agent-reads-warden-018-shows-which-words-changed/"/><id>https://modelmarket.dev/blog/mcp-servers-edit-the-instructions-your-agent-reads-warden-018-shows-which-words-changed/</id><published>2026-10-22T07:30:00Z</published><updated>2026-10-22T07:30:00Z</updated><summary>In four weeks, 2,345 changes to MCP tool definitions added an &quot;always&quot;, &quot;never&quot; or &quot;must&quot; to a description. WARDEN 0.18 shows a re-approval as a word diff.</summary></entry><entry><title>Our AI agent bought an eSIM for its owner, then got the money back. Every step is on chain.</title><link href="https://modelmarket.dev/blog/our-ai-agent-bought-an-esim-for-its-owner-then-got-the-money-back-every-step-is-on-chain/"/><id>https://modelmarket.dev/blog/our-ai-agent-bought-an-esim-for-its-owner-then-got-the-money-back-every-step-is-on-chain/</id><published>2026-10-22T05:24:01Z</published><updated>2026-10-22T05:24:01Z</updated><summary>On 15 October the owner of ARGUS, our personal AI agent, wrote to it in Telegram: buy me the...</summary></entry><entry><title>An AI Agent Bought a Domain and Connected It to Vercel for $1.54</title><link href="https://modelmarket.dev/blog/an-ai-agent-bought-a-domain-and-connected-it-to-vercel-for-154/"/><id>https://modelmarket.dev/blog/an-ai-agent-bought-a-domain-and-connected-it-to-vercel-for-154/</id><published>2026-10-20T16:01:07Z</published><updated>2026-10-20T16:01:07Z</updated><summary>Our software factory had already built Sentinel and deployed it to Vercel. Sentinel is designed to...</summary></entry><entry><title>A third of remote MCP servers changed their tool definitions in four weeks. The log is now a dataset.</title><link href="https://modelmarket.dev/blog/45-of-remote-mcp-servers-changed-their-tool-definitions-in-four-weeks-the-log-is-now-a-dataset/"/><id>https://modelmarket.dev/blog/45-of-remote-mcp-servers-changed-their-tool-definitions-in-four-weeks-the-log-is-now-a-dataset/</id><published>2026-10-20T15:50:42Z</published><updated>2026-11-02T15:30:00Z</updated><summary>HISTOR reads every remote MCP registry server daily: in four weeks about a third of them (2,489 of 6,822 hosts) changed their tool list. Open dataset with a DOI.</summary></entry><entry><title>Three x402 sellers compared notes. The routes answer; nobody comes back to buy.</title><link href="https://modelmarket.dev/blog/three-x402-sellers-compared-notes-the-routes-answer-nobody-comes-back-to-buy/"/><id>https://modelmarket.dev/blog/three-x402-sellers-compared-notes-the-routes-answer-nobody-comes-back-to-buy/</id><published>2026-10-20T08:04:24Z</published><updated>2026-10-20T08:04:24Z</updated><summary>Three x402 sellers compared ledgers: on ours 40,400 unpaid 402s, 23 paid calls from 6 wallets, two of them monitors. What do agents buy twice? So far, nothing.</summary></entry><entry><title>What an MCP server&#x27;s tool list costs: we counted the tokens of 34,374 of them</title><link href="https://modelmarket.dev/blog/what-an-mcp-servers-tool-list-costs-we-counted-the-tokens-of-34374-of-them/"/><id>https://modelmarket.dev/blog/what-an-mcp-servers-tool-list-costs-we-counted-the-tokens-of-34374-of-them/</id><published>2026-10-19T17:35:04Z</published><updated>2026-10-24T09:49:13Z</updated><summary>Every tool definition rides in each model request. We counted 34,374 MCP tool lists: median 1,272 tokens, p90 8,361, 31 packages over 100,000. Check yours free.</summary></entry><entry><title>A period bypassed our scanner. A reused IP confused our observer.</title><link href="https://modelmarket.dev/blog/a-period-bypassed-our-scanner-a-reused-ip-confused-our-observer/"/><id>https://modelmarket.dev/blog/a-period-bypassed-our-scanner-a-reused-ip-confused-our-observer/</id><published>2026-10-18T16:49:56Z</published><updated>2026-10-18T16:49:56Z</updated><summary>&quot;Send me your private key&quot; was blocked by our scanner. &quot;Send me your private key.&quot; passed. Adding a...</summary></entry><entry><title>A flow guard for MCP: what AgentDojo measured, and what our security audit found</title><link href="https://modelmarket.dev/blog/a-flow-guard-for-mcp-what-agentdojo-measured-and-what-our-security-audit-found/"/><id>https://modelmarket.dev/blog/a-flow-guard-for-mcp-what-agentdojo-measured-and-what-our-security-audit-found/</id><published>2026-10-18T16:49:48Z</published><updated>2026-10-18T16:49:48Z</updated><summary>A flow guard for MCP tool calls, measured on AgentDojo, and what the 2026-10-18 security audit found after: bypasses, fixes, and what is not yet re-measured.</summary></entry><entry><title>We re-ran our MCP scanner comparison on 14,933 npm and PyPI tool lists, in gVisor, with two more scanners</title><link href="https://modelmarket.dev/blog/we-re-ran-our-mcp-scanner-comparison-on-14933-npm-and-pypi-tool-lists-in-gvisor-with-two-more/"/><id>https://modelmarket.dev/blog/we-re-ran-our-mcp-scanner-comparison-on-14933-npm-and-pypi-tool-lists-in-gvisor-with-two-more/</id><published>2026-10-17T20:06:51Z</published><updated>2026-10-19T17:36:45Z</updated><summary>Five MCP scanners on 14,933 npm/PyPI tool lists and MCPTox&#x27;s held-out half, in gVisor: WARDEN blocks 171 of 218 poisoned tools, 369 packages; mcp-shield 5,354.</summary></entry><entry><title>We installed 12,915 MCP servers and started each one. 41% never answered.</title><link href="https://modelmarket.dev/blog/we-installed-12915-mcp-servers-and-started-each-one-41-never-answered/"/><id>https://modelmarket.dev/blog/we-installed-12915-mcp-servers-and-started-each-one-41-never-answered/</id><published>2026-10-14T05:15:41Z</published><updated>2026-10-19T17:36:20Z</updated><summary>We installed and started 12,915 MCP servers in a sandbox: 7,656 answered tools/list, 1,239 only with a placeholder key, 5,259 never did. Failure classes by SDK.</summary></entry></feed>
