Publish /.well-known/mcp/server-card.json — A Builder’s Guide
Robauto Team · Published · Last updated
Claude, ChatGPT, and Grok Bot connectors need a discoverable MCP server card. Here’s how to publish /.well-known/mcp/server-card.json and verify it with Robauto agent_scan’s mcp_card check.
If Claude, ChatGPT, or Grok Bot connectors cannot discover your tools, start with one file: /.well-known/mcp/server-card.json. Anthropic open-sourced the Model Context Protocol so assistants can attach to data and tools through a shared contract — and Robauto’s agent_scan checks whether that card is published.
Why builders need an MCP server card
On November 25, 2024, Anthropic introduced the Model Context Protocol (MCP) as an open standard for connecting AI assistants to systems where data lives. Docs at modelcontextprotocol.io describe the client/server model: expose tools through an MCP server; assistants connect as clients.
A public server card at a well-known URL turns private desktop config into something connectors can find without a human pasting an endpoint. Robauto’s Agent Integration check mcp_card looks for /.well-known/mcp/server-card.json.
What Robauto’s agent_scan checks
A live agent_scan of robauto.ai on 2026-09-19 scored 98/100 (pixel-verified). The MCP finding:
- id:
mcp_card - title: Share your tool context engine
- status: pass
- detail: MCP server card published — assistants can attach without a human pasting a URL.
- evidence: https://robauto.ai/.well-known/mcp/server-card.json
- spec: modelcontextprotocol.io
Tiers on that scan: Signal Basics 6/6, Technical Groundwork 4/4, Agent Integration 9/10 (95%), Agentic Commerce 5/5. Re-run any domain at robauto.ai/scan.
{
"example_scan": {
"domain": "robauto.ai",
"scanned_at": "2026-09-19T21:59:48.107Z",
"score": 98,
"tiers": {
"Signal Basics": "6/6",
"Technical Groundwork": "4/4",
"Agent Integration": "9/10",
"Agentic Commerce": "5/5"
},
"mcp_card": {
"status": "pass",
"evidence": "https://robauto.ai/.well-known/mcp/server-card.json",
"serverInfo.name": "robauto",
"transports": ["streamable-http → https://mcp.robauto.ai"]
}
},
"sources": [
"https://robauto.ai/scan?url=robauto.ai",
"https://robauto.ai/.well-known/mcp/server-card.json",
"https://www.anthropic.com/news/model-context-protocol",
"https://modelcontextprotocol.io"
]
}
Minimal server-card shape
Publish JSON at https://YOUR_DOMAIN/.well-known/mcp/server-card.json. Include serverInfo, transports, a honest tool catalog, and auth expectations matching your protected-resource metadata.
Builder checklist
| Step | What to ship | Why it matters | Source |
|---|---|---|---|
| 1. Spec literacy | Read MCP architecture | Avoid one-off connector glue | modelcontextprotocol.io |
| 2. Announce path | /.well-known/mcp/server-card.json over HTTPS | Connectors can discover you | Anthropic MCP announcement |
| 3. Declare transport | Stable endpoint + auth mode | Clients attach without a pasted URL | robauto.ai server card |
| 4. List tools honestly | Only tools you expose | Hallucinated tools waste agent turns | MCP docs |
| 5. Pair with OAuth discovery | /.well-known/oauth-protected-resource | Scored in Agent Integration | agent_scan example |
| 6. Verify with agent_scan | Confirm mcp_card = pass | Catch missing cards early | robauto.ai/scan |
Claude, ChatGPT, and Grok Bot connectors
MCP began with local servers and open SDKs. Remote well-known discovery lets product connectors fetch a card, negotiate transport, and call tools. Publish the open card first; map product-specific connector UIs onto the same endpoint.
Network context: Agent Readiness Index median is still 18 across 79 scored of 99 attempted (readiness index). An MCP card is high-leverage Agent Integration work.
Sources
- Anthropic: Introducing the Model Context Protocol
- Model Context Protocol documentation
- Robauto agent_scan: robauto.ai
- Robauto MCP server card
- Robauto Agent Readiness Index
Where can an agent get this data live?
Robauto MCP — Agent Intelligence & Brand Discovery Infrastructure. A public, stateless Streamable HTTP endpoint at https://mcp.robauto.ai/mcp, callable with no key. Setup and the full tool list are in the Robauto MCP directory.