
mercury-x402-mcp
Agent-payable web data over x402: 18 keyless services an agent pays per call.
What it can do
What data it sees
Do you need an account
No: the server works without sign-in
Agent-payable web data over x402: 18 keyless services an agent pays per call.
How to connect
How to connect mercury-x402-mcp to Claude
The server address to paste into Claude settings:
https://network.mercury-hq.com/mcp
- Open Claude (claude.ai or the desktop app).
- Go to Settings → Connectors.
- Click Add custom connector.
- Paste the server address copied below into Remote MCP server URL and click Add.
- In a chat, click + → Connectors and switch the new connector on.
Custom connectors are available on Free, Pro, Max, Team and Enterprise plans (Free is limited to one). On Team and Enterprise an organization Owner adds the connector first under Organization settings → Connectors.
Authorization
After you click Add, a sign-in window for the service opens. Sign in with your own account and approve access. Claude never sees your password.
How to connect mercury-x402-mcp to ChatGPT
The server address to paste into ChatGPT settings:
https://network.mercury-hq.com/mcp
- Open ChatGPT in a browser (chatgpt.com). A Plus, Pro, Business, Enterprise or Edu plan is required.
- Turn on developer mode once: Settings → Apps → Advanced settings → Developer mode.
- Open Settings → Connectors and click Create.
- Fill in the form: Name (anything), Description (one line about what the service does), MCP server URL (copy it below).
- Under Authentication choose OAuth if the service requires sign-in, otherwise None. Click Create.
- In a new chat open + → Apps/Connectors and enable the connector.
OpenAI has renamed this section before (Connectors → Apps/Plugins). If the label differs, search settings for "developer mode". On Business/Enterprise workspaces an admin must allow custom connectors first.
Authorization
On first use ChatGPT opens the service's sign-in window. Sign in and approve access.
How to connect mercury-x402-mcp to Cursor
The server address to paste into Cursor settings:
https://network.mercury-hq.com/mcp
Fastest: click Open in Cursor below and confirm the prompt.
Manually:
- In Cursor open Settings → Cursor Settings → MCP and click Add new global MCP server.
- Paste the JSON copied below into
~/.cursor/mcp.json(per project:.cursor/mcp.jsonin the repo root). - Save the file. The server appears in the MCP list; authorize it there if asked.
Authorization
If the service needs sign-in, an authorize button appears next to the server in the MCP list.
Server tool list (18)
Raw names from tools/list. Only developers need these.
| fetch | VERIFIABLE keyless web-read for autonomous agents. Every result ships a cryptographically SIGNED provenance receipt (EIP-191 over sha256(text)+url+status+time) — the wedge a free scraper structurally CANNOT match: Jina r.jina.ai is free+keyless too, but its bytes are HEARSAY (no proof of what/where/when). MERCURY's `attestation` is ecrecoverable OFFLINE, forever, by you OR any downstream agent you forward the bytes to — proving the content is genuine + untampered (key pinned at /.well-known/mercury-attestation). For RAG, trading and agent-to-agent commerce that need provenance, that is the gap between data and evidence. Beyond that it's the keyless web-read primitive — NO API key, NO signup, NO account, NO monthly plan, the one fetch SKU a fresh agent can onboard to by itself instead of stopping to ask a human for a key. Give a ?url= and get back clean readable page text + title + status. Agent-native extras (opt-in): ?format=markdown for structure-preserving markdown, ?links=1 for an outbound-link graph (crawl frontier), and the headline wedge — STRUCTURED EXTRACT: ?extract=title,price,author,publishedAt returns a clean JSON record { title, price, author, publishedAt }, an LLM-ready row not a wall of text. That is Firecrawl's paid 'JSON mode' (they need an LLM call + an API key for it) done here DETERMINISTICALLY from the page's own JSON-LD/OpenGraph/meta/microdata — keyless, no LLM, $0.003. (?extract=1 still returns the legacy description + wordCount.) The extracted record is folded into the SIGNED attestation too, so a buyer can prove the FIELDS — not just the raw bytes — are exactly what MERCURY resolved. You pay in-band over HTTP 402 (x402, USDC on Base mainnet) — the wedge those tools can't match: they ALL gate behind a human-created API key + a credit-card plan, so an agent can't onboard itself. This one an agent finds in the x402 Bazaar and pays with zero human in the loop. Honest charge-per-ATTEMPT: every call returns a structured result (success OR an ok:f |
| extract | TYPED structured extract for autonomous agents — URL + schema → a clean, type-safe JSON record. Where /buy/fetch returns page TEXT (and ?extract= returns string-only fields), THIS returns the schema-conformant object an LLM/RAG/trading pipeline actually consumes: pass ?url=…&schema=title,price:number,rating:number,inStock:boolean and get back { title:"…", price:19.99, rating:4.5, inStock:true } — numbers as numbers, booleans as booleans, absent fields null (honest). `schema` accepts the URL-friendly compact form (field[:type], type in string|number|integer|boolean) OR a Firecrawl/OpenAI-style JSON-Schema object ({"properties":{"price":{"type":"number"}}}). That is Firecrawl's paid 'JSON mode' headline guarantee — type-safety, 'numbers as numbers not strings' — done DETERMINISTICALLY from the page's own JSON-LD/OpenGraph/meta/microdata: keyless, NO LLM call, NO API key, NO signup, $0.004/call, paid in-band over HTTP 402 (x402, USDC on Base mainnet). The typed record is folded into the SIGNED provenance attestation too (EIP-191, ecrecoverable OFFLINE), so a buyer can prove the EXTRACTED FIELDS — not just raw bytes — are exactly what MERCURY resolved. Honest charge-per-ATTEMPT: every call returns a structured result (success OR an ok:false reason). Same SSRF guard, 5s timeout, 10MB cap, no mint. — $0.004/call |
| markdown | URL → clean, LLM-ready markdown (boilerplate/nav/ads stripped, headings + lists + links preserved) with a signed provenance receipt pinning the markdown to its source — the RAG-ingest primitive. Deterministic (no LLM): same URL + same source bytes ⇒ byte-identical markdown. — $0.005/call |
| metadata | URL → one typed, SIGNED metadata record (JSON-LD + OpenGraph + Twitter-card + standard <meta> + canonical + title), by source. Deterministic, keyless, no LLM — the social-card/SEO/schema.org record an agent can PROVE. — $0.006/call |
| links | URL → a SIGNED outbound/internal link graph: every <a href> as an absolute URL + anchor text, classified internal vs external against the page origin, grouped by origin with a third-party-origin histogram (privacy-audit) and same/cross-origin counts (SEO). Deterministic — same page bytes ⇒ byte-identical graph; no LLM. Covers <a> hyperlinks only (not rel/asset tags). Receipt = EIP-191 over the graph. — $0.005/call |
| robots | Domain → signed, timestamped per-AI-crawler allow/block audit from robots.txt + llms.txt + ai.txt (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bytespider, …). Deterministic, no LLM. EU-AI-Act / TDM opt-out evidence: the signed verdict proves the crawler policy as it stood at fetch time. — $0.005/call |
| diff | URL + prior content-hash (or prior text) -> refetch, deterministic change-proof: changed? + a NEW signed receipt binding fromHash->toHash. Stateless; receipts chain into a tamper-evident audit trail of a page's evolution. Keyless x402, Base mainnet USDC. — $0.006/call |
| notarize | Notarize any content (inline ?content= or a fetched ?url=) into a signed, offline-verifiable provenance receipt — sha256 contentHash + witnessed timestamp, attested by Mercury's pinned key. Deterministic, keyless, no LLM. The signed receipt is the product: it witnesses that these exact bytes existed in this exact form at this time (the one thing your own sha256() can't — a third-party witness). — $0.008/call |
| headers | URL → a deterministic HTTP security-headers audit (HSTS, CSP, X-Frame, X-Content-Type, Referrer-Policy, Permissions-Policy + more) with a letter grade and concrete findings, wrapped in a signed, offline-verifiable provenance receipt. Keyless, no LLM, no signup. — $0.005/call |
| table | URL in → the page's main HTML <table>(s) parsed into typed, header-keyed rows (JSON) + clean RFC-4180 CSV, with a signed provenance receipt over the exact extracted grid. Deterministic, keyless, no LLM — x402, USDC on Base mainnet. — $0.006/call |
| feed | RSS/Atom feed URL → a SIGNED, normalized item list [{title,link,published,summary}] unified across RSS 2.0/RDF + Atom 1.0: CDATA unwrapped, HTML stripped, entities decoded, dates → ISO-8601 (publishedRaw kept verbatim, never fabricated). Deterministic — same feed bytes ⇒ byte-identical items; no LLM. Core 4 fields only (not full-content/media/author). Receipt = EIP-191 over the canonical item list. — $0.005/call |
| availability | URL → a signed uptime/status probe: up/down, HTTP status + class, reachability reason, final URL after redirects, and a measured responseMs — wrapped in an offline-verifiable provenance receipt that makes "it was up/down at T" provable SLA evidence. Deterministic verdict (responseMs is telemetry, not signed). Keyless, no LLM, no signup. — $0.005/call |
| validate | URL (a JSON/API endpoint) + a JSON Schema in → a deterministic pass/fail with per-field errors (missing/wrong-type/enum/range/pattern), plus a signed provenance receipt binding the verdict to the exact response bytes AND the exact schema. Deterministic, keyless, no LLM — x402, USDC on Base mainnet. — $0.005/call |
| batch | List of URLs (≤20) → clean content for each + ONE signed receipt committing to a MERKLE ROOT over every page's contentHash. Tamper-evident multi-page snapshot with per-page membership proofs (selective disclosure). Deterministic (no LLM): same URL set + same source bytes ⇒ byte-identical root + proofs. SSRF-guarded per url; keyless x402, USDC on Base mainnet. — $0.02/call |
| sitemap | Domain/URL → a SIGNED snapshot of the site's PUBLISHED sitemap: discovers the sitemap via robots.txt Sitemap: lines then /sitemap.xml fallback, parses <urlset> + <sitemapindex> (follows up to 5 child sitemaps), returns a deduped, bounded (≤2000) URL inventory with lastmod/changefreq/priority. The receipt signs the DECLARED URL list (deterministic — same sitemap bytes ⇒ byte-identical list). Optional ?fetch=N (≤10) adds a HARD-BOUNDED same-domain liveness probe (title+status+bytes per URL) — that probe is the ONLY non-deterministic part and is NOT covered by the signature. SSRF-guarded; the crawl is bounded at every axis. — $0.01/call |
| dns | Domain → a SIGNED, timestamped DNS snapshot (A, AAAA, MX, NS, TXT, CNAME, SOA), normalised + sorted into a byte-stable record set with an offline-verifiable provenance receipt — proving exactly what the zone resolved to at that moment. Keyless, no LLM, no signup. (Normalisation is deterministic; DNS itself is mutable across time/TTL/geo — which is the very reason the snapshot is timestamped + signed.) — $0.006/call |
| readability | URL → a clean ARTICLE record { title, byline, publishedAt, article text } with boilerplate (nav/header/footer/sidebar/ads/share-bars/comment-forms) stripped via deterministic DOM density heuristics, plus a signed provenance receipt pinning the cleaned article to its source — the clean-citation primitive distinct from raw markdown. Deterministic (no LLM): same URL + same source bytes ⇒ byte-identical output. — $0.005/call |
| redirect | URL → a SIGNED redirect/canonical resolution: the full hop chain [{url,status}] from the link you have to where it ACTUALLY lands, the final resolved URL + status, the page's <link rel=canonical>, hop count, cross-origin flag and distinct origins traversed (affiliate-cloak / link-safety signal). Deterministic — same redirects ⇒ byte-identical resolution; no LLM. Follows HTTP 3xx only (no JS/meta-refresh execution — meta-refresh target surfaced un-followed). Receipt = EIP-191 over the resolved chain. — $0.005/call |