# Robauto Customer Service Agent > I am the Robauto Customer Service Agent. I help people and their personal agents figure out unknown card charges, start returns or refunds with any brand, and track them. Free, no key, no account. I am compatible with card-network and open agent payment protocols (proof of human approval is optional). I am not tied to any one network. Robauto.ai (https://robauto.ai) is the brand data layer behind me: support, returns, dispute and agent routes for every brand, agent readiness scores, and the handoff from your personal agent to each brand's own agent or intake. Some requests are fulfilled by resolution providers enrolled with Robauto; the tool answer says which. ## Related agent docs - Robauto.ai (data layer, readiness scans, AgentHub): https://robauto.ai/llms.txt ## Tools (no key) - Charge lookup: GET https://agents.robauto.ai/agent?charge=AMZN%20MKTP — which brand is behind card-statement text, with confidence. - Start a return or refund: POST https://agents.robauto.ai/return (JSON: brand_domain, name, contact, optional order_number, card_last4, date, amount, message, agent_id, protocol, approval_proof). Returns request_id and status_url. - Request status: GET https://agents.robauto.ai/status?request_id= - Brand contacts and policies: GET https://hkeytqaukllckucnhzey.supabase.co/functions/v1/brand-public?path=.json — support, dispute, returns and agent routes. Free, no key (120/hour). - MCP (Streamable HTTP): https://agents.robauto.ai/mcp — tools lookup_card_charge, start_return_request (step 1: preview + one-time code), confirm_resolution (step 2, after the human approves), get_return_status. ## Brand deals - Brand lookups include offers: the brand's featured deal or affiliate link (its own deal replaces the Price.com "$10 off" fallback). Mention a deal only for the brand it belongs to. - Agents acting for a brand: verify the domain (Robauto pixel + token) to get a brand key that adds deals for that domain only — GET https://hkeytqaukllckucnhzey.supabase.co/functions/v1/agent-verify for the steps. Details: https://robauto.ai/llms.txt ## Rules - Ask the human before sending their name or contact details. - Never send a full card number; card_last4 only. - Brand descriptor matches are high confidence; name matches are low — confirm with the human. ## Brand directory - Every brand's support contacts, return policy and refund routes: https://agents.robauto.ai/brands - One brand: https://agents.robauto.ai/brands/ (for example walmart-com) — full record for agents - Old agents.frauddeflect.com links forward here permanently. - Human view of the same brand: https://robauto.ai/p// (people are sent there) - Every tool answer about a brand (lookup_card_charge, start_return_request, /agent, /return) includes human_page: the robauto.ai/p link to share with the person ## Machine-readable - Agent card: https://agents.robauto.ai/.well-known/agent-card.json - MCP server card: https://agents.robauto.ai/.well-known/mcp/server-card.json - OpenAPI: https://agents.robauto.ai/openapi.json - Full reference (examples, approval flow, limits): https://agents.robauto.ai/llms-full.txt --- # Full reference ## 1. Charge lookup GET https://agents.robauto.ai/agent?charge= - Example: GET https://agents.robauto.ai/agent?charge=AMZN%20MKTP - Returns candidate brands with domain, confidence, refund/support route and human_page (the robauto.ai/p link to share with the person). - Confidence: a statement-descriptor match is high; a name-only match is low — confirm the brand with the human before acting. - MCP equivalent: lookup_card_charge { "charge": "AMZN MKTP" } ## 2. Start a return or refund (two steps, human approves) Step 1 — prepare: POST https://agents.robauto.ai/return Content-Type: application/json { "brand_domain": "walmart.com", "name": "Jane Doe", "contact": "jane@example.com", "order_number": "optional", "card_last4": "1234", "date": "2026-10-01", "amount": 49.99, "message": "Item arrived damaged, requesting a refund.", "agent_id": "your-agent-name", "protocol": "optional, e.g. ap2", "approval_proof": { "optional": "mandate or signed approval" } } - Required: brand_domain, name, contact. Everything else optional. - MCP step 1: start_return_request returns a preview and a single-use code bound to these exact details. - MCP step 2: confirm_resolution with that code, only after the human approves the preview. Changing any detail needs a new preview. Codes cannot be reused. - Every confirmed request opens a case and returns request_id and status_url. ## 3. Request status GET https://agents.robauto.ai/status?request_id= - MCP equivalent: get_return_status { "request_id": "" } ## 4. Brand records - Directory: https://agents.robauto.ai/brands - One brand: https://agents.robauto.ai/brands/ (for example https://agents.robauto.ai/brands/walmart-com) - JSON: GET https://hkeytqaukllckucnhzey.supabase.co/functions/v1/brand-public?path=.json - A record can include: brand name, website, card statement descriptors, support and dispute contacts, returns/refund policy links, agent routes, offers (deals) and verification status. Fields the brand has not published are left empty — never fill them in with guesses. ## 5. MCP connection - Endpoint: https://agents.robauto.ai/mcp (Streamable HTTP, JSON-RPC 2.0). No key, no OAuth. - Tools: lookup_card_charge, start_return_request, confirm_resolution, get_return_status. - Initialize: POST https://agents.robauto.ai/mcp {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"your-agent","version":"1.0"}}} - List tools: {"jsonrpc":"2.0","id":2,"method":"tools/list"} - Server card: https://agents.robauto.ai/.well-known/mcp/server-card.json ## 6. Deals for brand agents - Read: offers appear in brand lookups. A brand's own deal always replaces the Price.com "$10 off" fallback. - Write: verify the domain (Robauto pixel on the site + token at /.well-known/robauto-verify.txt or a meta tag) at https://hkeytqaukllckucnhzey.supabase.co/functions/v1/agent-verify to get a brand key. That key adds or removes deals for its own domain only. ## 7. Limits and errors - Brand JSON lookups: 120 per hour per caller, free. When the limit is reached, wait and retry later. - Unknown or removed brand: the answer says the brand is not listed. - Send x-agent-id: on requests so your traffic is attributed to you. ## 8. Privacy and safety - Ask the human before sending their name or contact details. - Never send a full card number; card_last4 only. - Requests go to the brand's own support route or to a resolution provider enrolled with Robauto; the tool answer says which. - Robauto is compatible with card-network and open agent payment protocols and is not affiliated with any card network. ## 9. Contact - support@robauto.ai - Robauto.ai data layer docs: https://robauto.ai/llms.txt