SameDayDesk x402 and MPP Data Gateway

Optional bounded batch extraction

Paid GET /extract and GET /read (and MCP extract / read) return a typed record after settlement. Merchant HTTP 200 is delivery of that record, not proof that the source succeeded or that the page is complete. The JSON keeps requestedUrl, observed finalUrl (also copied to url for compatibility), source HTTP status, sourceOk, a nullable error, and a capture object that labels the no-JavaScript GET, body/excerpt limits, charset, and truncation. A 403/404 with block text is not an empty 200. Observational HTTP delivery evidence may classify schema conformance, source refusal, and truncation; that is not buyer usefulness or a sold guarantee. The extract text field is a bounded excerpt. Omitted requests keep 1,200 characters. Optional textExcerptLimitChars from 1 through 40,000 is rejected before payment when it is not an integer in that range; the price stays the same and /read is not fetched. 40,000 matches the /read Markdown ceiling because the body was already capped at 3 MB. Batch requests use the same parser and a tighter ceiling so five excerpts cannot pass admission and then exceed the 128 KiB response cap. Use /read for longer Markdown. Neither route executes JavaScript, so missing discussion text is not proof of absence. Script data islands and comments are stripped; this is not a full-discussion reader. Production source connections validate and pin public IPv4 DNS answers, revalidate up to three redirects, and reject URL credentials, private addresses, IPv6 and compressed responses that ignore the requested identity encoding. Header/meta charset labels select a decoder; capture.charset names the decoder actually used. Unknown labels fall back to UTF-8 with an explicit fallback marker.

Batch rows retain requested source slots, including duplicates and failures. Each captured row's provenance.capture records the no-JavaScript method, actual byte allowance and 1,200-character text limit when text is requested. A known body or text cutoff is partial, not complete. This does not detect undisclosed missing content. Source refusal remains a failed row even when its error page has text. The in-process fetch/timeout test hooks have no HTTP, MCP or environment input binding and do not change settlement or authorization middleware.

EXTRACT_BATCH_ENABLED=1 adds POST /extract/batch and MCP extract_batch at an introductory 0.01 USDC per admitted batch of 1–5 public HTTPS URLs. The default is off: 25 paid HTTP operations and 22 MCP tools; enabled: 26 and 23. This price is not a measured margin guarantee. Use one server process and persistent COMMERCE_DATA_DIR; unknown settlement is quarantined, not automatically charged again. Bounded partial output does not promise every source or requested field.

LOCKFILE_PIN_DELTA_ENABLED=1 adds POST /lockfile-pin-delta and MCP lockfile_pin_delta for two caller-supplied npm package-lock.json objects. The default is off. Production already has EXTRACT_BATCH_ENABLED=1 (26 paid HTTP / 23 MCP tools). Enabling lockfile beside that envelope is 27 / 24. Default-off catalogs remain 25 / 22. Default price is $0.005 (LOCKFILE_PIN_DELTA_PRICE), matching live GET /extract. Margin is not proven. Initial live release accepts x402 only; existing MPP routes including extract_batch are unchanged. HTTP JSON only: no filesystem paths, arbitrary commands, or network fetch. Identical pins are informational, not engine failure. The signed deployment statement is not rewritten here. The lockfile MCP tool's challenge resource is https://agents.samedaydesk.com/lockfile-pin-delta, not mcp:// and not /extract/batch.

The extract_batch MCP tool projects the existing HTTP payment and durable replay handler. Its challenge resource is https://agents.samedaydesk.com/extract/batch, not mcp://: first call without payment, then reuse the returned HTTP resource and exact arguments with _meta["x402/payment"]. MPP callers may use the unmodified www-authenticate challenge in result _meta["samedaydesk/http"].headers and send its credential as HTTP Authorization. That metadata also carries the HTTP status and receipt headers. Never translate an MCP-resource credential. Existing MCP tools retain their existing native MCP payment boundary.

The signed statement declares 25 standard route capabilities in either flag state (Circle is separate); it does not claim that a disabled route is available or authorize payment. Live discovery excludes disabled batch extraction.

Every paid HTTP response advertises /.well-known/agent-payment-evidence.json through the standard HTTP describedby link relation. The bounded free manifest joins each exact method and route to its seller-declared read-only effect, recursively guaranteed JSON paths, response-schema digest, exact replay binding, receipt behavior, and the existing signed deployment statement. It does not change x402 or MPP terms and is not authorization to spend; buyers must still verify the live challenge, paid output, receipt, and settlement.

Twenty pay-per-call tools for deterministic agent-work opportunity preflight, machine-service discoverability, payment-offer preflight, Morpho borrower and market decisions, protection plans, URL extraction, Markdown reading, repository security scans, company and wallet enrichment, structured data generation, transaction receipt evidence, delegated-signer policy conformance, settlement proof, and AI-search readiness audits.

• Product page: https://samedaydesk.com/x402
• Smithery: https://smithery.ai/servers/epistemedeus/x402-data-gateway
• Remote MCP: https://agents.samedaydesk.com/mcp
• Agent Plugins 1.0 local package: plugins/samedaydesk-x402 (not marketplace-listed)
• Claude Code marketplace: .claude-plugin/marketplace.json plus plugins/samedaydesk-extract (samedaydesk-extract@samedaydesk-claude; not an Anthropic official directory listing)
• Goose native config: goose/ (YAML, session flag, and deeplink; not a Goose installer)
• Hermes Agent skills: hermes/ plus portable plugins/samedaydesk-x402/skills (AgentSkills SKILL.md drop-in; not a Hermes plugin or installer)
• Live resource manifest: https://agents.samedaydesk.com/.well-known/x402
• OpenAPI: https://agents.samedaydesk.com/openapi.json
• Official MPP OpenAPI: https://agents.samedaydesk.com/mpp-openapi.json
• Skill contract: https://agents.samedaydesk.com/skill.md
• Action catalog: https://agents.samedaydesk.com/api/actions
• A2A agent card: https://agents.samedaydesk.com/.well-known/agent-card.json
• Global A2A Registry: https://www.a2a-registry.org/agent/9cb0b8e6-cb1f-422b-a604-861d0a79e24b
• Settlement Radar: https://agents.samedaydesk.com/platforms
• Platform health JSON: https://agents.samedaydesk.com/v0/cards.json
• Aggregate machine-demand telemetry: https://agents.samedaydesk.com/v0/commerce-demand.json
• Morpho position risk: GET /defi/morpho-position?address=0x...&shocks=-10,-20,-30
• Morpho protection quote: GET /defi/morpho-protection?address=0x...&targetHealthFactor=1.25&protectAgainstShockPct=-10
• Morpho market underwriting: GET /defi/morpho-market-underwrite?marketId=0x...
• Morpho PreLiquidation replay: GET /defi/morpho-preliquidation-replay?transactionHash=0x...
• Opportunity preflight: GET /work/opportunity-preflight?rewardUsd=10&hours=0.25&hourlyCostUsd=4&selectionProbabilityPct=20
• Agent discoverability audit: GET /distribution/agent-discoverability-audit?origin=https://example.com&intent=extract+a+public+website+into+structured+JSON&route=/extract&method=GET&runtimeUrl=https%3A%2F%2Fexample.com%2Fextract%3Furl%3Dhttps%253A%252F%252Fexample.org&surfaceAudit=true&materializationAudit=true
• Payment offer preflight: GET /commerce/payment-offer-preflight?url=https://example.com/paid-route
• Seller integrity audit: GET /commerce/seller-integrity-audit?origin=https://seller.example&route=/paid-route&method=GET&requiredPaths=data.attributes
  • A buyer may add referral=r1_<sha256> derived from its complete seller-signed x402 settlement receipt. This creates a declared acquisition label, not verified proof. POST /commerce/referral-recheck accepts that receipt plus one downstream seller-signed settlement receipt from a distinct payer and transaction, then atomically grants at most one free changed-state recheck. The receipts prove neither an application HTTP 200 response nor output delivery or validity, and public posting or broadcasting is optional. A file matching Agent402 PR 1070's observable capture contract (bare pretty JSON receipt, no bundled Agent402 source) is already a drop-in input after ordinary JSON.parse; see docs/agent-outcome-referral-experiment.md and agent402-receipt-interop.test.mjs.
• Discovery↔live payment drift (library/CLI only; not a hosted or paid route): node discovery-drift.mjs observe --url 'https://agent-economy-signal-x402-mainnet.bronzetti-andrea.workers.dev/premium/agent-brief' --bazaar-pay-to '0xbda48b29607b9dc66ef7e38b68ad53f2b17efb23'
• Contract-qualified search: GET /commerce/contract-qualified-search?query=service+domain+ownership+code+provenance&requiredPaths=data.sourceRepository returns a bounded advisory OpenAPI repair plan for missing buyer-required paths without mutating seller files or inferring undeclared property types.
• Base USDC settlement proof: GET /commerce/settlement-proof?transactionHash=0x...&recipient=0x...&amountAtomic=5000
• Base or Ethereum transaction receipt: GET /chain/transaction-receipt?transactionHash=0x...&network=base
• Wallet policy conformance: POST /security/wallet-policy-conformance
• Wallet policy conformance contract: GET /schemas/wallet-policy-conformance-v1.json
• Stateful wallet policy conformance: POST /security/stateful-wallet-policy-conformance
• Stateful wallet policy contract: GET /schemas/stateful-wallet-policy-conformance-v1.json
• Material-change alert probe: https://agents.samedaydesk.com/alerts
• Agoragentic seller callback: POST /integrations/agoragentic/ai-readiness-audit
• the402 signed fulfillment webhook: POST /integrations/the402/webhook

No API key or subscription is required. Every paid HTTP route advertises x402 and native MPP Payment authentication in the same 402 response. Both protocols settle the same exact USDC amount to the same Base mainnet merchant wallet. MCP tool calls remain x402-gated.

The Morpho route is read-only. It calculates LTV, LLTV, health factor, liquidation headroom, and collateral-price shock scenarios from integer protocol values, then cross-checks indexed collateral, borrow shares, and oracle price against direct Base RPC state. Scenarios are calculations rather than probabilities or transaction recommendations. The separate protection route uses a fresh direct-RPC oracle read and direct confirmation of collateral and borrow shares to calculate exact partial-repay and add-collateral amounts. It returns unsigned token-approval and Morpho-call templates, explicit execution buffers, revalidation requirements, and economic postconditions. It never accesses a wallet, signs, broadcasts, or takes custody.

The service also keeps a privacy-safe demand telescope on a persistent Railway volume. It records route families, query key names whose runtime values pass the shared GET presence test, challenge/success classes, and pseudonymous repeat-use signals. External fetches are acquisition signals, not verified buyers, because unidentified automated indexers can remain. Recognized crawler and agent-indexer user agents are reduced at ingestion to a controlled source label and reported in a separate machine-discovery lens with source and route coverage. Those observations measure fetches, not authenticated catalog referrals, intent, or demand. Unmatched requests are reported separately from a conservative semantic-candidate subset; neither becomes demand until an independent caller repeats or converts. It does not record raw IP addresses, user agents, URLs, query values, bodies, payment headers, marketplace payloads, or credentials. Public output is aggregate only; owner traffic is excluded and crawler traffic remains excluded from demand even when its controlled discovery counts are reported separately. Common exploit probes such as .env, .git, and WordPress discovery paths are classified as scanner traffic and excluded as well.

Version 1.9.4 adds explicit paid-traffic classes without exposing buyer addresses. COMMERCE_PAYER_CLASSES accepts a JSON array of { "address", "class" } records. Controlled classes are internal, validation, incentivized, affiliated, and independent. Addresses are converted to the same secret-keyed payer pseudonyms already used by telemetry and classified at read time, which also permits retroactive correction without storing a raw address. Unknown payers remain unclassified; unfamiliar wallets never become independent demand by inference. The public snapshot reports paid success by class plus independent and repeat-independent actor counts.

Version 1.9.5 adds route-level paid-success counts inside each evidence class. This lets downstream monitors treat marketplace validation as accounting and transport evidence, alert on unclassified paid use for investigation, and advance the demand thesis only for explicitly independent or repeat-independent buyers.

Version 1.9.6 privately captures a valid Base transaction hash from successful x402 PAYMENT-RESPONSE or MPP Payment-Receipt headers after an explicit evidence baseline. Public telemetry exposes only proof coverage, distinct-count, and missing-reference counts by payment class. Raw headers and transaction references stay on the private volume. A missing reference becomes a material settlement-integrity event without exposing the reference itself.

Version 1.11.2 content-negotiates the root without changing its machine contract. Browser navigation with an explicit Accept: text/html receives a responsive human map of the fourteen tools, payment flow, and authoritative discovery links. JSON clients, curl's wildcard accept header, and agents keep the stable JSON descriptor. The response varies on Accept, and the human page duplicates no payment schema.

Version 1.11.2 also adds the source-attributed machine-discovery lens. It keeps raw user agents and network addresses out of the public snapshot, reports exact future indexer observations by controlled source and route, and gives the radar only first-source and first-route coverage changes as material events. Repeated crawl volume remains visible data without becoming an attention alert or demand.

Version 1.11.3 makes that reach lens prospective and self-excluding. A distinct COMMERCE_AGENT_DISCOVERY_SINCE baseline prevents pre-instrumentation crawler history from becoming attributed reach, while SameDayDesk-owned monitor user agents are excluded from both discovery and external-demand observations. This keeps integrity sweeps, brand-blind benchmarks, and radar probes from creating their own acquisition signal.

Version 1.11.4 preserves paying agents even when their user agent identifies as a crawler or indexer. A valid submitted x402 or MPP credential moves the event into economic telemetry before crawler classification, while the controlled user-agent label records source-to-paid conversion by source and route. Channel labels are still self-declared rather than authenticated referral proof, and independent demand still requires the explicit payer-class policy.

Version 1.11.5 separates paid-route reach from challenge delivery. Prospective agent/indexer observations now report paid-route probes, HTTP 402 challenges, distinct and repeat challenge actors, challenge rate, and controlled source and route breakdowns. This identifies whether the machine funnel stops before the paywall, at the challenge, or after a submitted credential without treating an indexer probe as purchase intent.

Version 1.11.6 adds a conservative challenge-to-payment cohort. A paid success counts as continuation only when the same secret-keyed network-and-user-agent actor returns after its first prospective challenge. The public snapshot reports converted calls, converted actors, independent converted actors, conversion rate, and controlled source and evidence-class totals without actor IDs. Network or user-agent drift can only create false negatives, so the metric is a lower bound rather than an identity claim.

Version 1.11.7 adds project-owned Glama connector verification at /.well-known/glama.json using the public SameDayDesk business email. Glama requests have their own controlled discovery-source label, so a propagated directory claim can be measured without becoming demand.

Version 1.11.8 starts a separate credential-attempt funnel. After a declared baseline, a parseable attempt must include a syntactically complete x402 v2 exact binding or MPP evm/charge credential. Signature validity and settlement remain later outcomes. Public aggregates separate header noise from parseable attempts and report protocol, result, route, controlled source, and explicit payer class without raw credentials, actor IDs, or addresses.

Version 1.11.9 improves MCP tool selection without renaming or duplicating any tool. Every tool now has a unique action-oriented title. The overlapping web and company tools explicitly say when to choose extract versus read, enrich versus schemaforge, and the combined deep_audit; wallet_enrich also states that its input is an EVM address rather than a company domain. The payment routes, names, prices, schemas, and handlers are unchanged.

Version 1.11.10 adds explicit descriptions to every opportunity_preflight input and to the three Morpho protection controls. This improves machine call construction while leaving names, routes, prices, required inputs, defaults, payment gates, and execution behavior unchanged.

Version 1.11.11 gives all four Morpho MCP tools explicit sibling-selection guidance. Borrower diagnosis, future protection planning, market underwriting, and historical PreLiquidation replay are now distinct machine choices without renaming a tool or changing its route, price, schema, payment gate, or handler.

Version 1.11.12 starts a prospective MCP transport-friction probe. It separates four common client expectations, /mcp/sse, /mcp/messages, /mcp/tools, and /mcp/events, from arbitrary /mcp/* misses without serving a guessed alias. Public aggregates expose only route counts and secret-keyed actor totals. A compatibility route is justified only by repeated independent use or conversion.

Version 1.11.13 repairs the pre-payment response contract exposed to machine buyers. All thirteen routes already authored explicit JSON output schemas, but the Bazaar v2 helper expects that schema under output.schema; the previous top-level outputSchema field was silently ignored. A single tested adapter now places each authored schema at the protocol-defined location, so an unpaid 402 challenge exposes both the example and the concrete required response fields before an agent authorizes payment. Routes, inputs, prices, settlement, and handlers are unchanged.

Version 1.11.14 projects the same thirteen response contracts into the free OpenAPI and action catalog. Discovery agents can now inspect concrete required fields and an example before probing a paid route; /read is also described as the JSON object its handler actually returns rather than a raw Markdown string. One route-keyed contract map drives the x402 challenge, OpenAPI, and action catalog to prevent the three machine surfaces from drifting apart.

Version 1.11.16 links the versioned, credential-free agent-payment-policy reference from the machine root, OpenAPI service metadata, and llms.txt. The reference has no wallet executor, payment signer, custody, or hosted paid verifier. It gives machine buyers a stable policy and evidence primitive without changing the merchant's routes, prices, payment requirements, or settlement.

Version 1.13.1 advances that machine-facing buyer reference to public package 0.4.0. The root, OpenAPI metadata, and llms.txt now advertise exact execution-shape authorization plus fourteen-dimension control-coverage schema v2. This follows first-person Tempo and Solana evidence that provider-native method and instruction allow rules can admit duplicated approved actions. No merchant route, price, handler, payment requirement, or settlement changes.

Version 1.14.0 turns the cross-chain delegated-signer failure we encountered into a credential-free paid product. POST /security/wallet-policy-conformance accepts only a bounded standardized allow, deny, and error matrix. It separates operation allowlisting from exact execution-shape control, credits only an explicit provider policy denial as provider-native enforcement, and returns conformant, partial, or unsafe with no opaque score. Invalid or secret-bearing shapes are rejected before payment. The evaluator accepts no wallet credential, signature, transaction body, wallet access, or broadcast authority and does not claim to have executed the caller's provider tests.

Version 1.14.1 publishes the exact standardized cases, evidence classes, input schema, output schema, method, protocols, and atomic price at the free canonical /schemas/wallet-policy-conformance-v1.json contract. The paid route, price, decision logic, payment gates, and settlement remain unchanged. Machine clients can now construct and validate the matrix without decoding a payment challenge.

Version 1.14.2 moves the taxonomy, strict validator, offline evaluator, and JSON Schemas into public agent-payment-policy@0.5.0; the hosted product now imports that package instead of maintaining a private duplicate. The public package also provides wallet-policy-init, wallet-policy-check, and wallet-policy-schema CLI commands. An unrun intended case is correctly partial, while a proven blocked intended action or an allowed mutation remains unsafe. The hosted route, 0.01-USDC price, payment terms, and credential-free boundary remain unchanged.

Version 1.14.3 advances the public standard dependency to agent-payment-policy@0.5.1. Provider-native control credit now requires every observed case for that control to pass. A denied optional shape case can no longer mask an allowed duplicate-approved-action case. The Privy Tempo and Solana adapters both classify exact execution shape as unverified and the overall native policies as unsafe, matching the first-person evidence. Route, price, payment, and credential boundaries remain unchanged.

Version 1.15.0 adds a separate stateful wallet-policy product from the project's first-person Privy cumulative-cap experiment. POST /security/stateful-wallet-policy-conformance evaluates seven safe standardized cases for sequential caps, signed-but-unbroadcast accounting, ABI extraction, concurrent oversubscription, counter-reference failure, and application serialization. GET /schemas/stateful-wallet-policy-conformance-v1.json publishes the free construction contract. The evaluator comes from public agent-payment-policy@0.6.0, accepts no credentials or raw provider payloads, and keeps provider-policy and application enforcement separate.