Setting Up Repliers MCP

Repliers MCP connects the Repliers API to AI assistants like ChatGPT, Claude, Claude Code and Codex, enabling you to leverage all of Repliers' features through natural language — including property searches, market analytics, listing history, and more.

There are two ways to use Repliers MCP:

• Hosted MCP (recommended) — connect directly to Repliers' MCP Server at https://mcp.repliers.io and sign in with your Repliers account. No server setup, and no API key ever goes into your client configuration.
• Open Source / Self-hosted (for technical users) — run your own instance using the Repliers MCP Server on GitHub with your API key in the environment. Recommended for developers who want full control over their environment or need a customized setup.

Step 1: Link Your API Key to MCP

Before connecting any MCP client to the hosted server, you need to authorize a Repliers API key for use with MCP. (Self-hosters skip this step — your key goes in a .env file instead, see Option B.)

1. Sign in to the Repliers Developer Portal
2. Find the API key you want to use and click the MCP icon next to it
3. That key is now linked to Repliers MCP

Step 2: Configure NLP for Listing Search

The Search_Listings tool works differently from most MCP tools — instead of relying on the AI assistant to structure the API request, it passes your natural language query directly to Repliers' own NLP endpoint. This approach was chosen because it produces significantly more accurate results: Repliers' NLP layer understands MLS-specific terminology, normalizes values to match MLS® standards, and supports context-aware conversational searches.

Because the NLP endpoint is powered by OpenAI, you must link a valid OpenAI API key to your Repliers API key before Search_Listings will work. Every other tool works without it.

How to Enable NLP Search

1. Sign in to the Repliers Developer Portal
2. Select the API key you have linked to MCP
3. Edit the key settings and locate the NLP Search option
4. Enter your OpenAI API key and save

Once configured, the Search_Listings tool will accept natural language queries and automatically convert them into structured Repliers API requests. Full guide: Utilizing AI-Powered NLP for Real Estate Listing Searches.

What NLP Search Can Do

• Natural language to structured queries — "Find me a condo in Toronto with at least 1 bedroom in the Annex with underground parking" is automatically converted into the correct Repliers API parameters
• Context-aware conversations — follow-up prompts refine the previous search without starting over (e.g., "Also, my budget is $500k and I need at least 1 parking spot")
• AI image search integration — visual preferences like "white kitchen" or "open concept living room" are incorporated into the search
• MLS® data normalization — user-friendly terms are mapped to valid MLS® values (e.g., "enclosed balcony" → balcony=encl)

How the assistant keeps searches honest

Every Search_Listings response leads with an appliedFilters block that shows which filters were actually applied — location, property type, style, price range, bedrooms and so on. The assistant is instructed to compare that block against your request before presenting results, because the NLP parser occasionally drops or substitutes a constraint. When it finds a mismatch it fixes it with refine-search, which re-runs the previous search with only the named parameters corrected, and can report the misparse to the Repliers team with send-feedback so the parser improves over time. You will always be told when a report is sent.

Option A: Hosted MCP Server (Recommended)

The hosted MCP server is the fastest and easiest way to get started — no installation or local server required. It runs at:

https://mcp.repliers.io

Both https://mcp.repliers.io and https://mcp.repliers.io/mcp are accepted; use whichever your client prefers. The server speaks Streamable HTTP and authenticates with OAuth 2.1: your client opens a browser window, you sign in with your Repliers account, and the client is returned an access token. The API key you linked in Step 1 is then used for every request — it never appears in your client's configuration.

Connecting via Claude (claude.ai and Claude Desktop)

1. In Claude, open Settings → Connectors
2. Choose Add custom connector
3. Enter the hosted MCP URL: https://mcp.repliers.io
4. Save the connector
5. When prompted, sign in to your Repliers account to authorize the connection

Once connected, Claude will have access to all Repliers MCP tools in your conversations. Connectors added on claude.ai are available in the Claude desktop and mobile apps too.

Connecting via ChatGPT

Before you start: remove any existing Repliers plugin. If you previously added Repliers to ChatGPT as a personal plugin (MCP server), you must remove it first. With two plugins offering the same tools, ChatGPT cannot decide which server to use and every request fails with an error.

1. In ChatGPT on the web, open Plugins and select Personal (https://chatgpt.com/plugins?view=personal). The custom MCP servers you added are listed under Created by me
2. Find your existing Repliers plugin, press … on its row and select Uninstall. If the row shows + instead of …, the plugin is already uninstalled — nothing to do

Then add the hosted server:

1. Still on Plugins → Personal, press the + button at the top right, next to the search box
2. Enter a name that is not already used by a plugin under Created by me — for example Repliers New if "Repliers" is taken. ChatGPT will not create a second plugin with a name that is already in use; the name is only a label and does not affect how the tools work
3. Enter the MCP server URL: https://mcp.repliers.io/mcp
4. Leave authentication set to OAuth and save
5. When prompted, sign in to your Repliers account to complete authorization
6. Review the list of discovered tools — you should see Search_Listings, Market_Statistics and the rest of the roster below

Watch The Video!

Our Co-Founder, Patrick Arlia recorded a video that shows you how to connect ChatGPT to Repliers. Click here to view it on YouTube.

Connecting via Claude Code

claude mcp add --transport http repliers https://mcp.repliers.io/mcp

Then run /mcp inside Claude Code and choose Authenticate next to repliers. A browser window opens for the Repliers sign-in; once it completes, the tools are available in your session.

Connecting via Codex, Cursor and other MCP clients

Add a remote (HTTP) server with the URL https://mcp.repliers.io/mcp, leave authentication set to OAuth, and complete the browser sign-in when the client prompts for it.

Supported MCP clients

The Repliers sign-in only returns to clients whose OAuth callback URLs are registered with us. These clients are supported today:

• Claude Desktop
• Claude Code
• ChatGPT
• Codex
• Copilot
• Cursor
• Antigravity
• Smithery
• Perplexity AI
• LeadConnector

If you connect from a client that is not on this list and get an OAuth error during sign-in, reach out to Repliers support — we can add it.

Option B: Open Source / Self-Hosted MCP Server (For Technical Users)

If you prefer to run your own MCP server — for example, in a production environment or to customize the toolset — you can use the open source version. This option is recommended for developers comfortable with the command line.

A self-hosted server has no user accounts or OAuth: it authenticates to Repliers with the API key in its environment, so it should only be exposed to people you trust with that key. It is ideal for personal use and internal tools.

Prerequisites

• Node.js — the version pinned in the repository's .nvmrc file (currently 26.x). The project enables engine-strict, so npm install refuses other major versions. If you use a version manager such as nvm or fnm, run nvm use / fnm use in the project directory first.
• npm (included with Node.js)
• A Repliers API key

Installation

1. Clone the repository

git clone https://github.com/Repliers-io/mcp-server.git
cd mcp-server

2. Install dependencies

npm install

3. Configure your API key

Create a .env file in the project root:

REPLIERS_API_KEY=your-repliers-api-key
PORT=3001

PORT only matters for the HTTP transport (below) and defaults to 3001.

4. Start the server

The server supports two transports from one entry point:

node mcpServer.js          # stdio — for Claude Desktop, Postman, and other local clients
node mcpServer.js --http   # Streamable HTTP on http://localhost:3001/mcp (and /) — for remote-MCP clients

Testing with Postman (Optional but Recommended)

Before connecting to an AI client, you can verify your server is working using the Postman Desktop App.

1. Download Postman Desktop
2. Create a new MCP Request
3. Set the type to STDIO and the command to:

node /absolute/path/to/mcpServer.js

To find the absolute path, run:

realpath mcpServer.js

4. Click Connect — you should see a list of available Repliers tools

Connecting to Claude Desktop

1. Get the absolute paths to your node binary and mcpServer.js:

which node
realpath mcpServer.js

2. Open Claude Desktop → Settings → Developers → Edit Config and add:

{
  "mcpServers": {
    "repliers": {
      "command": "/absolute/path/to/node",
      "args": ["/absolute/path/to/mcpServer.js"],
      "env": {
        "REPLIERS_API_KEY": "your-repliers-api-key"
      }
    }
  }
}

3. Save and restart Claude Desktop
4. Confirm the Repliers server has a green indicator under Settings → Developers

Connecting to Claude Code

claude mcp add --transport http repliers http://localhost:3001/mcp

with the server running as node mcpServer.js --http. No sign-in is needed for a self-hosted server.

Docker Deployment (Production)

For production use, Docker is recommended for reliability and isolation. One image serves both transports; the default command is Streamable HTTP.

1. Build the image

docker build -t repliers-mcp .

2. Run it

# Streamable HTTP (default) — endpoint http://localhost:3001/mcp (and /)
docker run --rm --env-file .env -p 3001:3001 repliers-mcp

# stdio — override the command, keep stdin open, no TTY, no port
docker run -i --rm --env-file .env repliers-mcp node mcpServer.js

3. For a stdio client such as Claude Desktop, point the client at docker:

{
  "mcpServers": {
    "repliers": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "--env-file", "/absolute/path/to/.env", "repliers-mcp", "node", "mcpServer.js"]
    }
  }
}

Make sure your .env file contains your REPLIERS_API_KEY. -i is required (without it stdin closes and the server exits); -t must not be set (a TTY corrupts the JSON-RPC stream). The image runs as the unprivileged node user and excludes .env, node_modules, .git, docs and test — configuration comes only from --env-file or -e.

Optional configuration

Restart the server after every .env change.

Available Tools

Once connected, your AI assistant can use the full suite of Repliers tools, organized by category:

Listings

Locations & Buildings

Agents, Members & Brokerages

Clients

Saved Searches

Favorites

Estimates

Messaging & NLP

Feedback

Every tool is annotated as read-only or mutating, so clients that hide write operations (for example read-only ChatGPT connectors) filter the list automatically.

Example Prompts

• "Find 3-bedroom apartments in San Francisco under $1 million listed in the last week"
• "Show me the cover photo of the first one"
• "Send listing alerts to my client john doe when 4 bedroom homes in austin hit the market"
• "Give me the listing history for 123 Main St, San Francisco"
• "What's the median list price for residential homes in Austin, aggregated by month over the last 18 months?"

Troubleshooting

The hosted MCP server isn't showing as connected
Make sure you've linked an API key in the Repliers Developer Portal (Step 1 above). Only one key can be active at a time. If you signed in but every tool call fails with "account not provisioned", the account you signed in with has no linked key.

I'm being asked to sign in again after an update
Expected once: the hosted server moved to OAuth 2.1 and existing connectors must re-authorize. Complete the sign-in and it will not ask again until the token expires.

Sign-in fails with an OAuth error (redirect or callback not allowed)
Your MCP client is probably not on the supported clients list — the Repliers sign-in only returns to registered callback URLs. Reach out to Repliers support with the name of the client the the required callback URLs and we can add it.

I'm being asked to sign in on every connection
Ensure you're completing the Repliers sign-in flow fully. If the issue persists, remove and re-add the connector, or unlink and re-link your API key in the Developer Portal.

Search_Listings fails but other tools work
NLP is not enabled on the linked API key. Follow Step 2.

ChatGPT shows no option to add a plugin, or the connection fails partway (desktop app)
Set up from the web app at https://chatgpt.com instead. The desktop apps do not currently support adding a custom MCP plugin; once it is created on the web, it appears in the desktop and mobile apps.

ChatGPT errors about which MCP server to use
An earlier Repliers plugin is still installed alongside the new one. In ChatGPT on the web, open Plugins → Personal, press … on the old Repliers plugin and select Uninstall, then try again — see Connecting via ChatGPT above.

ChatGPT will not accept the plugin name
A plugin with that name already exists under Plugins → Personal → Created by me — uninstalled plugins stay listed there and keep their name. Try … → Manage → Delete on the old plugin; if it is still listed afterwards (this does not always work), choose a different name, for example Repliers New. The name is only a label.

npm install refuses to run (self-hosted)
Your Node.js version does not match .nvmrc. Install the pinned version (or run nvm use / fnm use) and try again.

Claude Desktop falls back to an old Node version
Use the absolute path to your Node binary (from which node) rather than just node in your config.

Tools aren't appearing in Claude
Restart Claude Desktop after editing the config file. Confirm the server has a green status indicator under Settings → Developers. For the hosted connector, check that the connector is enabled for the current conversation.

My client connects but reports the wrong endpoint (self-hosted)
The HTTP transport is served at /mcp (and /) on the configured PORT. The old --sse flag now starts the same Streamable HTTP server.

Additional Resources

• Repliers Developer Portal
• Repliers MCP Server on GitHub
• Enabling NLP Search
• Repliers API Documentation
• Create a Repliers Account

Updated on: 14/09/2026