Web Search Research x402 offerings

Bulk detect page language: up to 20 urls in one call, processed concurrently, results in input order with a per-item error field and a failure count. Same answer per item as the single endpoint. Batch enrichment for agents that hold a list. $0.10 per batch.

Change watch for detect page language: runs the check, compares it with the result from your previous call for the same url (kept 90 days) and returns the current answer, changed true/false, the fields that changed with before/after values, and when it was last seen. Schedule it to be alerted when this changes. $0.02 per check.

Detect the page's last modified: fetches the URL and answers this one question from markup, headers and text, with the evidence. Cheap classification signal for crawlers and content pipelines. $0.01 per page.

Bulk detect page last modified: up to 20 urls in one call, processed concurrently, results in input order with a per-item error field and a failure count. Same answer per item as the single endpoint. Batch enrichment for agents that hold a list. $0.10 per batch.

Bulk detect page page type: up to 20 urls in one call, processed concurrently, results in input order with a per-item error field and a failure count. Same answer per item as the single endpoint. Batch enrichment for agents that hold a list. $0.10 per batch.

Change watch for detect page page type: runs the check, compares it with the result from your previous call for the same url (kept 90 days) and returns the current answer, changed true/false, the fields that changed with before/after values, and when it was last seen. Schedule it to be alerted when this changes. $0.02 per check.

Detect the page's paywall: fetches the URL and answers this one question from markup, headers and text, with the evidence. Cheap classification signal for crawlers and content pipelines. $0.01 per page.

Bulk detect page paywall: up to 20 urls in one call, processed concurrently, results in input order with a per-item error field and a failure count. Same answer per item as the single endpoint. Batch enrichment for agents that hold a list. $0.10 per batch.

Change watch for detect page paywall: runs the check, compares it with the result from your previous call for the same url (kept 90 days) and returns the current answer, changed true/false, the fields that changed with before/after values, and when it was last seen. Schedule it to be alerted when this changes. $0.02 per check.

Detect the page's rtl or ltr: fetches the URL and answers this one question from markup, headers and text, with the evidence. Cheap classification signal for crawlers and content pipelines. $0.01 per page.

Bulk detect page rtl or ltr: up to 20 urls in one call, processed concurrently, results in input order with a per-item error field and a failure count. Same answer per item as the single endpoint. Batch enrichment for agents that hold a list. $0.10 per batch.

An agent that evaluates a website needs every independently accepted URL finding from one shared bounded raw-HTML measurement.

x402 Services & Facilitator Registry — 2026-10-04 edition: The x402 market, measured, refreshed for October: every resource in Coinbase's x402 Bazaar discovery index (27,681 paid endpoints, up 87% in five weeks) with the index's own 30-day counters (settled calls, unique payers, last call), one read-only probe per URL (live 402 or not, version, live price), an EIP-712 signing-domain check on every USDC-on-Base option, per-resource revenue proxy, payTo addresses, networks, schemes and prices — plus a month-over-month changes file (new and vanished domains, biggest gainers and losers), 21 facilitators with live /supported results and the curated 118-service table. JSONL + JSON + schema + README. Market sizing, competitor research, facilitator choice, pricing and listing validation from one file.

MCP Server Registry Snapshot — 2026-08-22 edition: A curated, machine-readable snapshot of 400 production/community Model Context Protocol (MCP) servers as of 2026-08-22: category, transport, auth requirements, repo/package URLs, source-directory cross-references, and verification dates for every entry. Saves an agent or builder shopping for MCP tooling the work of crawling and reconciling multiple MCP directories (official registry, Smithery, Glama, PulseMCP, mcp.so) itself.

Prompt-Injection & Tool-Hijack Test Corpus: 156 structured, labeled adversarial test cases across seven categories (direct injection, indirect injection, tool hijacking, data exfiltration, system-prompt leaks, jailbreak/roleplay, encoding obfuscation) for evaluating whether an agent or LLM app resists prompt injection and tool-misuse attacks. Ships with a JSON Schema, a dependency-light Python evaluation harness, and a guide covering taxonomy, scoring, limitations, and responsible use.

Prompt-Injection Defenses Playbook & Detection Ruleset: The defense counterpart to our attack corpus: a layered playbook for hardening an agent/LLM app against prompt injection, tool hijacking, data exfiltration, system-prompt leaks, jailbreaks, and encoding obfuscation. Includes a 6-layer defense taxonomy mapped to attack categories, 119 machine-readable detection rules (85 base + 34 extended, each with honest false-positive notes), and a stdlib-only Python detector with two profiles (strict for tool arguments and output, content for web pages and documents), decode-and-rescan for obfuscated instructions, and hidden-HTML-text scanning. Measured: held-out recall 53% strict / 38% content on the 156-case corpus; 0.35% false positives on 1,714 unseen top-site homepages (the base rules as written flag 69%). Full method and limits in MEASUREMENTS.md. Grounded in OWASP LLM/MCP guidance.

MCP Live Tool Inventory — 2026-08-28 edition: Every remote server in the official MCP registry (15,329 URLs) probed read-only via initialize → tools/list, and everything the 8,235 live ones returned: 140,284 tool names, titles, full descriptions, input-schema keys and annotations; 5,462 server-level instructions strings verbatim; per-server reachability, protocol version and outcome; and 24k+ tool-poisoning audit flags with snippets. JSONL + summary.json + schema. The real-world base rate for allow-listing servers, name-collision checks, and building or testing tool-description detectors against live text instead of synthetic samples.

Agent Eval Harness Templates: A ready-to-use starter kit for evaluating an agent or LLM application fast: a JSON-Schema test-case format, 40 example cases across four suites (task completion, tool selection, refusal/safety, regression), a pointwise LLM-as-judge scorer with documented bias mitigations, and a dependency-light Python CLI runner that produces a scored JSON report plus a human-readable summary. Grounded in current agent-eval practice; bring-your-own-model via a one-file adapter interface.

The Agent Web, Measured — 2026 Q3 edition: A 25-page analyst report on the machine-facing web, built entirely from first-hand measurement: all 15,329 remote MCP registry URLs probed (54% answer, 45.5% outside two template fleets; 140,284 tool descriptions scored — the textbook tool-poisoning payload appears zero times, 47 hosts carry concealment directives, all 51 quoted verbatim), all 14,820 x402 Bazaar endpoints probed and the index's own counters summed into the first published revenue ceiling for the CDP-facilitated x402 market (~$314/day), all 21 facilitators probed, and a 60-crawler census from a live storefront. 22 charts, 10 tables, twelve conclusions, every number generated from the underlying datasets. PDF + README + licence. Single-organisation licence; includes the +30/+60/+90-day re-probe summaries and the 2026 Q4 edition.

x402 Quickstart Kit — 2026-08-23 edition: A dependency-minimal reference implementation of the x402 HTTP micropayment protocol ("exact" scheme, EIP-3009, EVM/USDC): seller-side payment middleware for Cloudflare Workers/Hono and plain Node/Express, generalized from a live production x402 seller, plus buyer-side client code in TypeScript (viem) and Python (eth_account/web3.py) that signs a real EIP-3009 payment authorization and retries a 402 automatically. Includes a practical guide covering facilitator selection, common integration pitfalls, and a go-live checklist.

Search the live web and get ranked results back: title, URL and snippet for each hit, up to 20 per call, optionally biased to a country. The natural first step before fetching or crawling - this tool finds the pages, the fetch and crawl tools read them. Results come from an independent web index and the response names which one answered.

Crawl a website: fetch a page, follow the links inside the same host and return every page it reached, within a page budget you set. For reading documentation, a product catalogue or a changelog in one paid call instead of many.

Fetch any public web page or JSON API and get its status, content type, headers and body back. For agents that need to read a page they cannot reach themselves. Private networks and loopback addresses are blocked, and large bodies are truncated rather than dropped.

Lists the pages of a website for a migration check: reads its sitemap or crawls from the start URL, and maps every page onto the target site when one is given. Use it before starting a comparison job to choose which pages to compare. 0.05 USD per call.