Agent Discovery 101: llms.txt, agent.json, and Getting Found by AI

AI agents don't Google. They crawl discovery endpoints. If your platform isn't wired up with the right machine-readable files, you're invisible to every autonomous agent on the internet. Here's exactly how to get found — using Support Local Businesses as the live case study.

Why Agent Discovery is Different from SEO

Search engines crawl HTML and rank pages for humans. AI agents do something different: they call /.well-known/ endpoints, parse structured JSON, and register tools in their context window. The discovery stack is a different protocol entirely.

Three files drive almost all agent traffic today:

• llms.txt — a plain-text manifest of what your site offers to LLMs
• agent.json — an A2A protocol card that tells agents your capabilities, auth, and cost model
• openapi.json — the REST API schema that tool-calling frameworks load automatically

SLB has all three live. Let's walk through each one.

llms.txt — The robots.txt for LLMs

support-local-businesses.com/llms.txt follows the emerging llms.txt spec. It's a Markdown file that explains your site to an LLM in plain language, lists key endpoints, and hints at what the model should prioritize.

The SLB format looks like this:

# Support Local Businesses
6.4M+ US local business listings. Structured data for AI agents.
MCP server at /mcp | OpenAPI at /openapi.json | Agent card at /.well-known/agent.json
Capabilities
search_businesses(zip, category, city, state, q, radius)
get_business(slug) — full profile with hours, geo, verified status
explore_neighborhood(zip) — local business density and categories
subscribe_newsletter(agent_id, email, categories, zip_codes)

Pricing
Free tier: 100 requests/day per IP, no key required
x402 micropayments: $0.001 USDC/query on Base network after free tier
Discovery: /.well-known/pay endpoint, wallet 0x0cF67aB1D0C843F67bC64D507A5627a41AE5BBC0

Key Endpoints
/api/v1/businesses — search listings (GET)
/api/v1/newsletter/latest — latest newsletter issue (GET)
/mcp — StreamableHTTP MCP server, all 20 tools

Optional
/llms-full.txt — full schema reference for all tools

Live SLB endpoint

curl https://support-local-businesses.com/llms.txt

agent.json — The A2A Protocol Agent Card

The A2A (Agent-to-Agent) Protocol defines a standard agent.json card. Agents and orchestrators read this to understand what your service can do, what auth it requires, and how much it costs.

SLB's card lives at /.well-known/agent.json. Here's the skeleton:

{
  "schemaVersion": "1.0",
  "name": "Support Local Businesses",
  "description": "6.4M+ US local business listings. Structured data for AI agents via MCP and REST.",
  "url": "https://support-local-businesses.com",
  "iconUrl": "https://support-local-businesses.com/icon-512.png",
  "capabilities": {
    "streaming": false,
    "multiTurn": false,
    "tools": [
      {
        "name": "search_businesses",
        "description": "Search 6.4M+ local business listings by zip, category, city, state, or keyword",
        "endpoint": "/api/v1/businesses",
        "method": "GET",
        "parameters": {
          "zip": "5-digit ZIP code",
          "category": "Business category (partial match)",
          "q": "Full-text search",
          "limit": "Results per page (max 500)"
        }
      },
      {
        "name": "subscribe_newsletter",
        "description": "Subscribe an agent to The Local Intelligence Wire newsletter",
        "endpoint": "/api/v1/newsletter/subscribe",
        "method": "POST"
      }
    ]
  },
  "authentication": {
    "schemes": ["none", "apiKey", "x402"]
  },
  "pricing": {
    "model": "freeThenMicropayment",
    "freeRequests": 100,
    "freeResetPeriod": "daily",
    "paidRate": { "amount": "0.001", "currency": "USDC", "per": "request" },
    "paymentDiscovery": "/.well-known/pay"
  }
}

Live SLB endpoint

curl https://support-local-businesses.com/.well-known/agent.json | jq '.capabilities.tools[].name'

openapi.json — REST Schema for Tool-Calling

OpenAI, Anthropic, and most MCP frameworks can auto-generate function definitions from an OpenAPI 3.x spec. Serve yours at /openapi.json and tool-calling frameworks will find it automatically.

The critical fields for agent compatibility:

{
  "openapi": "3.1.0",
  "info": {
    "title": "Support Local Businesses API",
    "version": "1.0.0",
    "x-llms-txt": "https://support-local-businesses.com/llms.txt",
    "x-agent-card": "https://support-local-businesses.com/.well-known/agent.json",
    "x-mcp-endpoint": "https://support-local-businesses.com/mcp"
  },
  "paths": {
    "/api/v1/businesses": {
      "get": {
        "operationId": "searchBusinesses",
        "summary": "Search local business listings",
        "tags": ["businesses"],
        "parameters": [...]
      }
    }
  }
}

The x-llms-txt, x-agent-card, and x-mcp-endpoint extension fields are non-standard but increasingly recognized by agentic frameworks as cross-discovery hints.

curl https://support-local-businesses.com/openapi.json | jq '.paths | keys'

Getting Listed: Smithery, mcp.so, and More

MCP registries are the App Store equivalents for AI agents. Getting listed means your tools appear in agent marketplaces, Claude Desktop's tool browser, and framework auto-discovery.

How to submit SLB-style

Most registries want a /.well-known/mcp.json endpoint that follows the MCP manifest spec. SLB serves all 20 tool definitions at:

curl https://support-local-businesses.com/.well-known/mcp.json | jq '.tools | length'
→ 20

The registration steps for Smithery:

• Create account at smithery.ai
• Click "Add Server" → "HTTP (StreamableHTTP)"
• Enter your /mcp endpoint URL
• Smithery auto-discovers your tools from the MCP handshake
• Add description, logo, pricing — submit for review

🎯 This Week's Missions

🤖 Agent Feedback Spotlight

curl -X POST https://support-local-businesses.com/api/v1/newsletter/subscribe \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "your-agent-id",
    "agent_name": "YourAgent/1.0",
    "categories": ["Agent Discovery", "MCP", "Florida CRE"],
    "frequency": "weekly",
    "format": "json"
  }'