Well-Known Agent Files

Detects the newer generation of .well-known machine-readable agent descriptors: an MCP server card (SEP-1649), an A2A agent card, or an agent-skills manifest. These let an agent discover, without a human in the loop, that a site exposes an MCP endpoint, an A2A-compatible agent, or a packaged set of skills it can use.

What it checks

The audit probes, once per crawl, for:

• MCP server card: /.well-known/mcp/server-card.json, plus the variants /.well-known/mcp.json, /.well-known/mcp, and /.well-known/mcp-server.
• A2A agent card: /.well-known/agent-card.json.
• agent-skills manifest: /.well-known/agent-skills/index.json.

For each, it reports presence, whether the body parses as valid JSON, and a summary of declared capabilities where the schema makes that straightforward. It also checks for /.well-known/ai-plugin.json, the deprecated OpenAI plugin manifest format: if present, it’s flagged as stale — the plugin system it described was retired, and a lingering manifest is more likely to confuse a modern agent than help it.

Solution

If the site backs an MCP server, an A2A-compatible agent, or a packaged skill set, publish the corresponding manifest at its .well-known path so agents can discover it without prior knowledge of your API:

{
  "name": "example-mcp",
  "version": "1.0.0",
  "description": "MCP server for Example Co's API",
  "endpoint": "https://example.com/mcp"
}

Remove any leftover /.well-known/ai-plugin.json from the ChatGPT-plugins era — it’s dead weight now, not a discoverability signal.

Enable / Disable

Disable this rule

[rules]
disable = ["ax/well-known-agent"]

Disable all Agent Experience rules

[rules]
disable = ["ax/*"]

Enable only this rule

[rules]
enable = ["ax/well-known-agent"]
disable = ["*"]