MCP server

ListingsFeed speaks the Model Context Protocol over streamable HTTP, so Claude, ChatGPT, Cursor, Windsurf, Claude Code and any MCP-capable agent can search the live dataset with the same key you use for the REST API.

Endpoint	`https://listingsfeed.com/mcp`
Transport	Streamable HTTP (JSON responses; no SSE stream required)
Auth	`Authorization: Bearer <api key>`
Protocol version	`2025-06-18` (older clients negotiate down)
Manifest	/.well-known/mcp.json

Tools

`count_listings`	Count matches for any filter set. Free. Use it first to size a query.
`search_listings`	Page through listings (default 25 per call, `limit` up to your plan's page size). Metered on records returned. Returns `next_cursor`.
`get_listing`	One listing by `hash_id` with every field. 1 record.
`list_brokerages`	Brokerages with active listings, by state or name. Free.
`get_stats`	Counts by state / type / transaction, fill rates, last refresh. Free.

Tool arguments mirror the REST filters (state, property_type, transaction, min_sf, near + radius_mi, updated_since, q…). Results come back as JSON text plus structuredContent.

Claude Desktop / Claude.ai

Settings → Connectors → Add custom connector: name ListingsFeed, URL https://listingsfeed.com/mcp. When asked for authentication choose a bearer token and paste your key. Or in claude_desktop_config.json via a stdio bridge:

{
  "mcpServers": {
    "listingsfeed": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://listingsfeed.com/mcp", "--header", "Authorization: Bearer ${LF_KEY}"],
      "env": {"LF_KEY": "lf_live_…"}
    }
  }
}

Claude Code

claude mcp add --transport http listingsfeed https://listingsfeed.com/mcp \
  --header "Authorization: Bearer lf_live_…"

Cursor / Windsurf / VS Code

{
  "mcpServers": {
    "listingsfeed": {
      "url": "https://listingsfeed.com/mcp",
      "headers": {"Authorization": "Bearer lf_live_…"}
    }
  }
}

ChatGPT

ChatGPT connects to remote MCP servers through custom connectors (Settings → Connectors → Create) and through the Apps SDK. Use the URL above with your bearer key. For a custom GPT without MCP, import /openapi.json as an Action with "API key → Bearer" auth.

Your own agent (Python)

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async with streamablehttp_client("https://listingsfeed.com/mcp",
        headers={"Authorization": f"Bearer {LF_KEY}"}) as (read, write, _):
    async with ClientSession(read, write) as session:
        await session.initialize()
        n = await session.call_tool("count_listings", {"state": "TX", "property_type": "industrial"})
        page = await session.call_tool("search_listings", {"state": "TX", "property_type": "industrial",
                                                            "min_sf": "100000", "limit": 25})

Raw JSON-RPC

curl -s -X POST https://listingsfeed.com/mcp \
  -H "Authorization: Bearer $LF_KEY" -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"count_listings","arguments":{"state":"OR","property_type":"land"}}}'

Good prompts

"How many industrial listings over 100,000 SF are for sale in Texas right now, and who are the top brokerages?"
"Find land over 200 acres within 50 miles of Columbus, OH with an asking price, and list the brokers."
"What changed in Phoenix office listings since last Monday?"

Metering is identical to the REST API: search_listings and get_listing bill records returned; counting, brokerages and stats are free. Quota errors come back as tool errors with the quota object, so the assistant can explain them.