4 min read

Using the Augrented MCP API: setup and authentication

How to connect AI coding agents (Claude Code, Hermes Agent, OpenClaw, ChatGPT) to the Augrented building data MCP server — setup, authentication, and tool reference for NYC, SF, LA, and Austin.

Updated Jun 18, 2026

Using the Augrented MCP API: setup and authentication

Category: Getting Started
Read time: 6 minutes

The Augrented building data MCP (Model Context Protocol) server lets AI coding agents look up building profiles, search violation records, inspect permits, analyze landlord portfolios, and check rent-regulation status across four supported cities — without leaving your development environment.

Instead of switching between the Augrented web interface and your editor, your agent can search buildings, fetch records, and compare landlords as native tool calls. This article covers how to connect each major AI coding tool to the Augrented MCP server.

Supported cities

The MCP server provides read-only access to building data in four cities:

City ID Identify by
New York City nyc BBL (Borough-Block-Lot)
San Francisco sf Block-Lot
Los Angeles la APN (Assessor Parcel Number)
Austin austin TCAD Parcel ID / GEO ID

Server details

The Augrented MCP server runs as a Streamable HTTP server at:

https://app.augrented.com/mcp

It exposes nine tool categories shared across all cities:

Tool group What it does
auth_status Check authentication status of your MCP session
select_city Set the target city for this session
find_buildings Search buildings by address fragment
get_building_profile Get a full building risk profile
get_building_records Fetch records from a specific dataset
find_landlords Search owner/management company names
select_landlord Pick a specific landlord for browsing
list_landlord_buildings Get a landlord's portfolio ranked by violations
get_selected_building Return the currently selected building

Authentication

The Augrented MCP server requires oAuth authentication - make sure you're logged into the site and you will be prompted to allow access when the tool loads in your agent.

Connection guides by client

Claude Code / Claude Desktop

HTTP (recommended):

Add the server to your project's .mcp.json:

{
  "mcpServers": {
    "augrented-buildings": {
      "type": "http",
      "url": "https://app.augrented.com/mcp"
    }
  }
}

Or use the CLI:

claude mcp add-json augrented-buildings '{"type":"http","url":"https://app.augrented.com/mcp"}'

Restart Claude Code. When authentication is required, Claude will detect the OAuth challenge and open a browser for the login flow.

Scope options: - local (default) — private to you, saved in ~/.claude.json - project — shared with the team via .mcp.json in the repo - user — available across all your projects

Hermes Agent

Add to ~/.hermes/config.yaml:

mcp_servers:
  augrented-buildings:
    url: "https://app.augrented.com/mcp"
    timeout: 120

Also accepts per-server tool filtering:

mcp_servers:
  augrented-buildings:
    url: "https://app.augrented.com/mcp"
    timeout: 120
    tools:
      include: [find_buildings, get_building_profile, get_building_records]

Reload with /reload-mcp in a Hermes session, or restart the client.

OpenClaw

Add via the CLI:

openclaw mcp add augrented-buildings \
  --transport streamable-http \
  --url https://app.augrented.com/mcp \
  --scope project

Or add to your OpenClaw config directly:

{
  "mcpServers": {
    "augrented-buildings": {
      "transport": "streamable-http",
      "url": "https://app.augrented.com/mcp"
    }
  }
}

Reload with openclaw mcp reload.

ChatGPT (Chat mode)

ChatGPT connects to MCP servers through the Connectors feature (ChatGPT Plus, Team, Pro, or Enterprise required):

  1. Open Settings → Connectors
  2. Under Advanced, toggle Developer Mode on
  3. Click Create to add a new connector
  4. Fill in:
  5. Name: Augrented Buildings
  6. Server URL: https://app.augrented.com/mcp
  7. Check I trust this application
  8. Click Create

If the server is in required auth mode, ChatGPT will handle the OAuth flow automatically — you'll be prompted to authorize through your browser.

In Chat mode, you must explicitly enable the connector per chat session (click + → More → Developer Mode and toggle the Augrented connector on).

Deep Research mode also supports MCP connectors — add the same URL and ChatGPT will use search and fetch tools from the server for research queries.

Verifying the connection

Once your agent connects, try:

Search building at 123 Main Street in NYC

If the MCP server loads correctly, your agent should respond by using the find_buildings or nyc_find_buildings tool, returning matching addresses with risk summaries.

To check auth status explicitly:

Check your Augrented MCP session auth status

This calls auth_status and returns whether you're authenticated, plus any available identity info.

Available city-specific tools

Each city registers its own set of prefixed tools:

City Prefix Sample tools
NYC nyc_ nyc_find_buildings, nyc_get_building_profile, nyc_find_landlords
SF sf_ sf_find_buildings, sf_get_building_profile, sf_list_landlord_buildings
LA la_ la_find_buildings, la_get_building_profile, la_find_landlords
Austin austin_ austin_find_buildings, austin_get_building_profile, austin_get_building_records

Generic tools (find_buildings, get_building_profile) work across all cities by first selecting the target city via select_city.

Tips

  • Start with public mode for development — no authentication needed until you're ready to add user identity tracking
  • Use generic tools for multi-city agents — find_buildings and get_building_profile handle city selection internally
  • Test server reachability before diving into agent configuration — curl https://app.augrented.com/mcp should return valid MCP metadata
  • Time out after 30 seconds on slow datasets (Austin 311 joins, LA historical permit lookups) — configure a timeout: 120 on your server entry

Ready to look up a building? Search any address on Augrented — free for your first 10 buildings.