MCP x402 - Evidence-Backed Web Verification

io.github.Diogoup26Published by GitHub account Diogoup26v1.2.12Updated Nov 1, 2026

Evidence-backed pre-purchase checks for offer terms, seller trust and policy compliance.

VerifiedStreamable HTTPWeb executableWeb Search & ScrapingFinanceAI & ML

Overview

AI-generated overview

Remote MCP server that checks a public web page against 1–10 conditions and returns evidence-backed confirm/reject/uncertain decisions, paid per call in USDC.

What you can do
  • Verify conditions on one public page: get confirmed, rejected or uncertain per condition with quoted evidence, source URL, timestamp, verification ID and page hash
  • Analyze a single public HTTP(S) page: source-bounded report with summary, facts, risks and recommended actions
  • Ask general questions: concise model answer without web retrieval or fact checking
  • Preview the result contract for free: static examples for offer terms, seller trust or policy compliance with no fetch, model call or payment
When to use it
For agents that must decide whether an offer, warranty, return policy, seller identity or policy term is supported by a page before buying or acting. Useful when a documented, quotable evidence trail matters more than a free-form summary.
Requirements
Remote streamable HTTP endpoint; no account, API key or header declared for MCP clients. Paid tools require an x402 payment of 0.02–0.05 USDC on Base mainnet, so the calling agent needs a funded wallet and x402-capable client. Self-hosting needs Node.js, an OPENAI_API_KEY and optionally OBSERVABILITY_SALT.
Before you install
Three of four tools cost money per call (consultar_ia 0.02 USDC; analisar_url and verificar_condicoes 0.05 USDC) on Base mainnet, so an agent can spend funds autonomously. The bundled local buyer needs EVM_PRIVATE_KEY, a wallet private key that must never be committed or exposed. Submitted URLs and page content are sent to the operator and to OpenAI; paid responses advertise a feedback endpoint.

Installation

In SourceWeft

  1. Open MCP x402 - Evidence-Backed Web Verification in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Use with an agent

Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.

Read https://sourceweft.com/mcp/io.github.Diogoup26%2Fmcp-x402/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.

Read the install guide the agent follows

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "x402-evidence-backed-web-verification": {
      "type": "http",
      "url": "https://mcp-x402-production.up.railway.app/mcp"
    }
  }
}

README

[MCP x402 logo]

MCP x402 — Evidence-Backed Web Verification for AI Agents

[AllMCPs Verified]

Verify a seller, product, offer, policy, or claim before an agent acts

Give an AI agent a public HTTP(S) page and 1 to 10 concrete conditions. The verificar_condicoes tool returns an evidence-based decision for every condition:

  • confirmado — the page clearly supports the condition.
  • rejeitado — the page clearly contradicts the condition.
  • incerto — the available page evidence is insufficient.

Each verification includes the final source URL, page title, timestamp, condition-by-condition explanation, quoted evidence when available, a unique verificationId, and a SHA-256 pageHash.

Designed for

  • AI purchasing and procurement agents;
  • offer, warranty, returns and availability checks;
  • seller identity and authorization checks;
  • policy and terms compliance checks;
  • commercial automation that requires evidence before taking action.

Three buyer decisions

Buyer needSelect this tool when the agent must decide
Offer termsWhether an offer satisfies warranty, returns, price or delivery requirements before purchase
Seller trustWhether the page identifies the seller and supports authorization or registration claims
Policy complianceWhether terms or policies contain required cancellation, privacy or support conditions

Use the free preview_verification MCP tool or the free HTTP endpoint POST /preview/verify-conditions first when the buyer needs to inspect the complete result contract without fetching a page, calling OpenAI or making a payment. Use verificar_condicoes for a fresh verification of a real public URL.

Simple x402 integration

  • Free preflight validation before payment.
  • $0.05 USDC per verification on Base mainnet.
  • No account or subscription required.
  • Available through MCP and HTTP.
  • Ready-to-run buyer included in this repository.
  • Bazaar discovery metadata enabled on the paid HTTP and MCP surfaces.

Free static result demonstration

The following example is fictitious and performs no request, OpenAI call, or payment. It shows the complete result contract an agent receives for 0.05 USDC when using verificar_condicoes.

Example input:

json
{  "url": "https://shop.example/offers/refurbished-laptop",  "condicoes": [    "The offer includes a 12-month warranty.",    "The return period is at least 30 days.",    "The seller is an authorized manufacturer reseller."  ],  "contexto": "Pre-purchase seller verification."}

Example result:

json
{  "source": "https://shop.example/offers/refurbished-laptop",  "title": "Refurbished Laptop Offer — Example Seller",  "verifiedAt": "2026-01-01T00:00:00.000Z",  "decisao": {    "decisao": "incerto",    "condicoes": [      {        "condicao": "The offer includes a 12-month warranty.",        "estado": "confirmada",        "prova": "Includes a 12-month limited warranty.",        "explicacao": "The offer explicitly states the warranty duration."      },      {        "condicao": "The return period is at least 30 days.",        "estado": "rejeitada",        "prova": "Returns accepted within 14 days of delivery.",        "explicacao": "The stated 14-day return period contradicts the required minimum of 30 days."      },      {        "condicao": "The seller is an authorized manufacturer reseller.",        "estado": "incerta",        "prova": null,        "explicacao": "The page provides no evidence of manufacturer authorization."      }    ],    "resumo": "The warranty is confirmed, the return-period requirement is rejected, and reseller authorization remains uncertain."  },  "verificationId": "00000000-0000-4000-8000-000000000000",  "pageHash": "sha256:0000000000000000000000000000000000000000000000000000000000000000"}

This demonstrates the distinction between supporting evidence, contradictory evidence, and absence of evidence. The real paid result uses the fetched page, current timestamp, a unique verificationId, and the SHA-256 hash of the page content.

Free MCP preview tool

preview_verification is a free, model-callable MCP tool. It performs no live web request, OpenAI call or payment. Choose one of these values:

  • offer_terms
  • seller_trust
  • policy_compliance

It returns a realistic structured input and result, identifies the paid verificar_condicoes tool and provides the exact next step. Preview calls are logged separately and never counted as paid demand or revenue.

Live endpoints

  • MCP: https://mcp-x402-production.up.railway.app/mcp
  • Free MCP preflight: POST https://mcp-x402-production.up.railway.app/preflight/mcp
  • Free verification preflight: POST https://mcp-x402-production.up.railway.app/preflight/verify-conditions
  • Free static HTTP preview: POST https://mcp-x402-production.up.railway.app/preview/verify-conditions
  • Paid verification: POST https://mcp-x402-production.up.railway.app/verify-conditions
  • Health: GET https://mcp-x402-production.up.railway.app/health

The service also provides paid public URL analysis and general AI consultation through analisar_url and consultar_ia.

Live Service

  • Base URL: https://mcp-x402-production.up.railway.app
  • Health check: GET /health
  • MCP endpoint: POST /mcp
  • Paid URL analysis: POST /analyze
  • Price: $0.05 USDC per paid request
  • Network: Base mainnet (eip155:8453)

Production Status

  • Production deployment active on Railway
  • x402 version 2 payment flow
  • Base mainnet payments (eip155:8453)
  • Paid HTTP endpoints return valid x402 challenges with Bazaar metadata
  • MCP and HTTP payment, execution, settlement, and delivery confirmed on-chain
  • MCP Bazaar metadata advertises the public HTTPS MCP server URL
  • CDP native MCP catalog indexing is not yet claimed as confirmed; upstream tracking: coinbase/cdp-sdk#764

MCP Tools

Decision verification

verificar_condicoes — $0.05 USDC

A paid decision-verification tool for agents.

Give it a public URL and one or more concrete conditions. It returns a decision based only on evidence extracted from that page:

  • confirmado — the page clearly proves the condition.
  • rejeitado — the page clearly contradicts the condition.
  • incerto — the page does not provide enough evidence.

Each result includes the final source URL, page title, verification timestamp, condition-by-condition explanation, and a short quoted proof where available.

Example conditions:

  • “The page identifies the seller.”
  • “The product is available in Portugal.”
  • “The page shows a price in euros.”
  • “The document states that the service supports x402.”

This is designed for agents that need an evidence-based decision before taking the next action.

ToolDescriptionPrice
preview_verificationReturns a static structured example for offer, seller or policy verification. No network retrieval or OpenAI call.Free
consultar_iaSends a prompt to the OpenAI Responses API.$0.02 USDC
analisar_urlFetches and analyzes a public HTTP or HTTPS page.$0.05 USDC
verificar_condicoesVerifies concrete conditions on a public page and returns evidence-based decisions.$0.05 USDC

Technology

  • Node.js and TypeScript
  • Express
  • Model Context Protocol
  • OpenAI Responses API
  • x402 payments
  • viem
  • Cheerio
  • Railway

Security

  • HTTPS provided by Railway
  • Helmet security headers
  • Rate limit of 100 requests per 15 minutes
  • Structured request logs with unique request IDs
  • Host validation and Railway health-check support
  • Restricted OpenAI API key with access only to /v1/responses
  • Environment files and secrets excluded from Git

Environment Variables

Server

VariableRequiredDescription
OPENAI_API_KEYYesOpenAI project API key.
OPENAI_MODELNoOpenAI model. Defaults to gpt-5-mini.
HOSTNoListening host. Defaults to 0.0.0.0.
PORTNoListening port. Defaults to 3000.
RAILWAY_PUBLIC_DOMAINRailwayAutomatically supplied by Railway.
PUBLIC_SERVICE_URLNoCanonical public HTTPS origin used in discovery metadata. Railway derives it automatically from RAILWAY_PUBLIC_DOMAIN; the production URL is the fallback.
OBSERVABILITY_SALTRecommendedStable secret salt used only to pseudonymize source/client fingerprints across restarts.

Local Buyer

VariableRequiredDescription
EVM_PRIVATE_KEYYesPrivate key of the Base mainnet buyer wallet. Never commit this value.
JOURNEY_IDNoExisting correlation ID to reuse; otherwise the buyer generates one.

Example local .env.test file:

env
EVM_PRIVATE_KEY=your_private_key_here

Local Development

Install dependencies:

bash
npm install

Start the development server:

bash
npm run dev

Build the project:

bash
npm run build

Start the compiled server:

bash
npm start

Health Check

The health endpoint is public and does not require payment:

bash
curl https://mcp-x402-production.up.railway.app/health

Railway Log Diagnostics

Railway currently accepts --since for historical HTTP logs but rejects --until. On Windows PowerShell, export a bounded application and HTTP log window without using the broken flag:

powershell
.\scripts\export-railway-window.ps1 `  -StartUtc "2026-08-22T16:31:22.401228537Z" `  -EndUtc "2026-08-23T06:21:18.045229877Z" `  -OutputPrefix "railway-diagnostic"

The script retrieves logs from the lower bound and applies the upper UTC bound locally using ordinal ISO-8601 comparison, which preserves Railway's nanosecond timestamps. It writes raw captures plus bounded app and http NDJSON files.

Analyze both bounded files while reading and validating every physical line:

bash
npm run logs:analyze -- \  --since "2026-08-22T16:31:22.401228537Z" \  --pretty \  --out railway-diagnostic-report.json \  railway-diagnostic-app.ndjson railway-diagnostic-http.ndjson

The report includes per-file coverage, empty/valid/invalid line counts, timestamp bounds, separation of Diogo-*, RailwayHealthcheck, known probes or indexers and potentially external traffic, journey reconstruction and funnel stopping points. It joins application railwayRequestId values to Railway HTTP requestId values, understands HTTP clientUa, groups server-generated journey IDs by fingerprint and temporal proximity, and counts only accepted normalized feedback. It deliberately does not treat an isolated 402 as purchase intent and does not infer human identity or motivation from request metadata.

Ready-to-Run x402 Buyer

The repository includes ready-to-run buyers for applications and AI agents. The MCP buyer connects to the live service, validates the x402 payment requirements against Base mainnet USDC and a maximum payment of $0.05, signs the payment, retries the same tool call, and prints the settlement receipt and result.

Clone and prepare the buyer:

bash
git clone https://github.com/Diogoup26/mcp-x402.gitcd mcp-x402npm install

Create a local .env.test file in the project root:

env
EVM_PRIVATE_KEY=0xYOUR_64_HEX_CHARACTER_PRIVATE_KEY

Never commit .env.test or expose the private key.

Build the project:

bash
npm run build

The following commands authorize real x402 payments on Base mainnet:

  • consultar_ia: $0.02 USDC
  • analisar_url: $0.05 USDC
  • verificar_condicoes: $0.05 USDC

HTTP URL Analysis

bash
npm run analyze -- "https://example.com" "Summarize this page."

HTTP Decision Verification

bash
npm run verify -- "https://example.com" "The page identifies the seller."

MCP URL Analysis

bash
npm run mcp:analyze -- "https://example.com" "Summarize this page."

MCP AI Consultation

bash
npm run mcp:consult -- "Reply only with: MCP OK"

MCP Decision Verification

bash
npm run mcp:verify -- "https://example.com" "The page identifies the seller."

A successful MCP payment prints:

  • PAGAMENTO MCP: success
  • the settlement receipt
  • the tool response

The MCP buyer never prints the private key and rejects any payment that:

  • is not on Base mainnet;
  • does not use the configured Base USDC contract;
  • exceeds $0.05 USDC.

Machine-Readable Payment Continuation

The discovery document at GET /.well-known/x402 and every successful free preflight response publish a complete continuation sequence. A compatible agent no longer has to infer what to do after the first unpaid request:

  1. run the free preflight and keep its x-journey-id;
  2. send the validated request without a payment signature;
  3. read and verify the payment-required requirements;
  4. authorize only the expected network, asset, amount and recipient;
  5. retry the identical request with the x402 payment payload;
  6. verify both delivery and the settlement receipt.

REST metadata recommends @x402/fetch; MCP metadata recommends @x402/mcp. Paid responses also link back to the discovery instructions through x-payment-instructions. They additionally publish an RFC 8288 Link header with the registered payment and help relations, so generic HTTP agents can discover the same machine-readable continuation and optional feedback path without depending on project-specific headers. The flow never requests a private key, Railway or GitHub credentials, or an OpenAI key.

Optional Conversion Feedback

Every paid HTTP or MCP response advertises x-feedback-endpoint and the allowed normalized reason, stage, and intent values. The free preflight responses include the same information in a feedback block. Feedback is submitted to POST /feedback, contains no free text, and never includes a private key, prompt, URL, page content, or payment payload.

The three buyers accept:

text
--feedback-reason <research_only|no_wallet|unsupported_network|insufficient_funds|spending_not_authorized|price|insufficient_value|integration_error|other>--feedback-stage <discovery|preflight|payment|execution|delivery>--feedback-intent <research|analyze_page|verify_conditions|general_question|evaluate_service|other>

When --feedback-reason is supplied without --feedback-stage, the stage defaults to payment and the buyer does not authorize a payment. The HTTP buyers first receive the unsigned 402 challenge, submit the explicit feedback, and stop. The MCP buyer stops after its free preflight and submits the explicit feedback without requiring a wallet.

Example: report that the price stopped an evaluation, without paying:

bash
npm run analyze -- "https://example.com" "Summarize this page." --feedback-reason price --feedback-intent evaluate_service

To submit explicit feedback after a completed delivery, select that stage:

bash
npm run mcp:consult -- "Reply only with: MCP OK" --feedback-reason other --feedback-stage delivery --feedback-intent general_question

Buyers automatically submit only integration_error, and only for an objectively detected technical failure such as failed discovery/preflight, an HTTP 5xx result, a paid MCP tool error, or a missing settlement receipt after a payment was made. They never infer price, research_only, no_wallet, or another human or commercial motivation.

Controlled Smoke Test (No Payment)

After deployment, validate discovery, preflight, the OpenAPI feedback contract, normalized feedback submission, HTTP 402 challenges and feedback headers, HTTP method handling, MCP initialization, tool discovery and the MCP x402 challenge:

bash
npm run buildnpm run smoke

Use SERVICE_URL to target another deployment. The test uses one persistent x-journey-id, sends User-Agent: Diogo-Smoke/1.2.9, verifies that the MCP x402 challenge advertises the public HTTPS endpoint with type=mcp and the correct toolName, and never creates or signs a payment.

The three payment buyers now perform discovery and a free preflight before the paid request, using the same persistent journey ID throughout. Set JOURNEY_ID to reuse an existing journey; otherwise each buyer creates and prints one. The MCP buyer additionally requires the exact advertised price, Base USDC, Base mainnet, and the configured service recipient before it can sign.

Funnel Observability

The server emits structured events for MCP tool attempts, x402 challenges, payment verification, execution, settlement and final tool outcome. It does not log tool arguments or payment payloads. Settlement logs include the public network and transaction identifier so completed purchases can be counted and deduplicated.

Rejected MCP requests include safe protocol diagnostics such as method, content type, Accept, JSON-RPC shape and the SDK error classification. All events include the request and journey correlation fields when available. Calls to a known paid tool that never reach its payment wrapper additionally emit mcp_pre_payment_rejection, classified as invalid method, content type, JSON-RPC envelope, arguments, transport/protocol rejection, or an otherwise unreached handler. Argument values and payment payloads are never logged. OpenAI usage events inherit the same request, journey, client, and source correlation. Railway's X-Railway-Request-Id, edge POP, and request-start time are also recorded, while client source fingerprints use Railway's stable X-Real-IP value and remain pseudonymized with OBSERVABILITY_SALT.

Deployment

The main branch is connected to Railway. Every successful push triggers a new deployment. Railway uses:

bash
npm run buildnpm start

The deployment health-check path is /health.

License

ISC

Source: README.md at commit 11d4881

Tools

4

preview_verification

Preview an evidence-backed verification for free

Use this free static preview before paying for verificar_condicoes. Choose offer_terms to inspect warranty, returns and availability; seller_trust to inspect seller identity and authorization; or policy_compliance to inspect required policy terms. It performs no web request, OpenAI call or payment and returns a realistic structured example plus the exact paid next step. PT: Pré-visualização gratuita e estática da verificação.

ParametersuseCase

consultar_ia

General AI consultation

Use for general questions that do not require live web retrieval. Returns a concise model answer and does not fetch URLs or verify current facts. Use analisar_url for a source-bounded page report and verificar_condicoes for evidence-backed decisions. PT: Use para perguntas gerais sem pesquisa web em tempo real.

Parameterspromptrequired

analisar_url

Analyze a public web page

Use to fetch one public HTTP or HTTPS page and return a source-bounded report with summary, facts, risks and recommended actions. The optional objective controls focus but is never evidence. It does not consult other sources; use verificar_condicoes for condition-by-condition decisions. PT: Analisa uma página pública sem usar fontes externas.

Parametersurlrequiredobjetivo

verificar_condicoes

Verify conditions with evidence

Use before an agent trusts, buys or acts to test 1 to 10 explicit conditions against one public HTTP or HTTPS page. Best fits offer and warranty checks, seller identity or authorization checks, and policy compliance checks. Returns confirmed, rejected or uncertain for every condition with textual evidence when available, plus source, verifiedAt, verificationId and pageHash. Missing evidence produces uncertain, not a false confirmation. Use preview_verification first when the buyer needs a free example. PT: Verifica condições com prova auditável.

Parametersurlrequiredcontextocondicoesrequired

Version history

3
  1. v1.2.12LatestNov 1, 2026
  2. v1.2.10Nov 1, 2026
  3. v1.2.9Sep 16, 2026

Score

90% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 daysRequired25 pts
  • Auth declaredRequired10 pts
  • Tools listedRequired15 pts
  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.