Give your agent a live view of every payment.

The server is a thin process that translates MCP tool calls into authenticated REST calls against api.topropay.com. Nothing about your account leaves the boundary you already trust — the same scopes, the same audit trail, the same idempotency guarantees apply. The only difference is the caller is a model.

• Inspect

  Read live payments & routing

  Pull any payment, see every routed attempt, and reason over the decision the engine made. Read-only by default — no chance of moving funds.

• Analyse

  Approval-rate diagnostics

  Ask the agent why approval is down on a market or BIN. The MCP server returns connector scores, latency, and decline reasons in structured form.

• Draft

  Prepare refunds & captures

  Agents draft refund and capture proposals. A signed approval step from a human runs them — the agent never has unilateral write access.

• Reconcile

  Match settlement to orders

  Stream the reconciliation feed into the agent's context. It pairs deposits with order IDs and explains every variance.

• Compose

  Generate code & runbooks

  Because the schema is in the agent's context, it can author idempotent SDK calls, dashboards, and operator runbooks that compile first try.

• Audit

  Signed agent trail

  Every MCP call is logged with the agent identity, the prompt that triggered it, and a SHA-256 of the response — exportable for compliance.

{
  "mcpServers": {
    "topropay": {
      "command": "npx",
      "args": ["-y", "@topropay/mcp-server@latest"],
      "env": { "TOPROPAY_API_KEY": "sk_live_..." }
    }
  }
}

{
  "mcpServers": {
    "topropay": {
      "command": "npx",
      "args": ["-y", "@topropay/mcp-server@latest"],
      "env": { "TOPROPAY_API_KEY": "sk_live_..." }
    }
  }
}

{
  "mcpServers": {
    "topropay": {
      "command": "npx",
      "args": ["-y", "@topropay/mcp-server@latest"],
      "env": { "TOPROPAY_API_KEY": "sk_live_..." }
    }
  }
}

{
  "mcpServers": {
    "topropay": {
      "url": "https://mcp.topropay.com/sse",
      "headers": { "Authorization": "Bearer sk_live_..." }
    }
  }
}

1. 01Grab an MCP-scoped key from the dashboard (Settings → API → MCP).
2. 02Paste the JSON above into your client's config file.
3. 03Restart the client. The topropay server shows up under connected tools.
4. 04Ask the agent "summarise yesterday's approval rate by market". It calls approval.metrics and answers from real data.

• Resource topropay://schema/openapi Live OpenAPI 3.1 spec of the REST API — the agent reasons over field shapes.
• Resource topropay://schema/events Webhook event catalogue with every payload shape and emit condition.
• Resource topropay://docs/routing Plain-language explanation of routing strategies, with worked examples.
• Prompt diagnose_approval_drop Pre-built prompt that walks the agent through an approval-rate regression analysis.
• Prompt reconcile_settlement Match a settlement batch to the orders that funded it and surface anomalies.

Pass the key through the client's env block (local) or as a bearer header (remote). Never hard-code a key into a prompt — the agent will surface it back into context.

• Read-only default. Fresh keys can only read; write scopes are explicit and dashboard-gated.
• Drafts, not writes. Refunds and captures are staged. A human signs off in the dashboard before money moves.
• Per-agent identity. The MCP client name and version are sent on every call and stamped into the audit log.
• Argument cap. payments.search caps at 1,000 rows per call. Agents cannot exfiltrate the whole ledger in one prompt.
• Redacted PII. Cardholder PAN and CVC never leave the boundary; tokenised references do.
• Killswitch. Revoke a key from the dashboard and every active session drops on the next call.

• "Find every EUR payment from yesterday with status=failed, group by issuing BIN, and tell me which BINs cascaded but still declined."

  → payments.search + routing.explain. Returns a ranked list of issuers with their cascade-recovery rate.

• "Approval on DE Visa is down 3pp this week vs last. Why?"

  → approval.metrics with a weekly window and the diagnose_approval_drop prompt. Returns a narrated diagnosis.

• "Draft a refund for order A-9421 — customer says the second charge was a mistake."

  → payments.search on metadata.order_id + refunds.draft. Returns a proposal and a dashboard URL for sign-off.

• "Match yesterday's settlement batch to orders and flag anything that doesn't line up."

  → reports.download + the reconcile_settlement prompt. Returns a structured variance table.