MCP Server

Why MCP?

If you're building anything where an LLM needs UK property data (investor agents, lead-gen widgets, due-diligence copilots, valuation chatbots), MCP is the fastest path. Instead of writing REST integration code, you give the LLM a single connection URL and it figures out which tool to call.

How it works

Once you've connected PMI to your AI assistant, it works like this:

No new dashboard, no new login, no second invoice. If you can use Claude or ChatGPT, you can use PMI through MCP.

What you get

• 28 tools: one per REST endpoint, including all Serviced Accommodation tools.
• Pass-through auth: your existing PMI API key works unchanged.
• Same billing: MCP tool calls deduct from the same monthly credit allowance as REST calls. They appear in /v1/account/usage under the same endpoint names.
• Same rate limit: your tier's per-10-second cap applies whether the request comes via REST or MCP.
• Streamable HTTP: modern transport supported by Claude Desktop, Cursor, ChatGPT, and the official MCP SDKs in Python and TypeScript.

Tools available (28)

Same names and credit cost as the REST endpoints. When the LLM picks one, it sees:

What you can ask

Once PMI is connected, you stop reaching for the dashboard. You just ask the AI you already use. Here are the kinds of questions that work well:

See it in action

If you've never set up an MCP server before, these public walkthroughs from other providers show what the experience looks like. Most installs are 30 seconds and one config-file edit. The PMI setup is identical: same JSON shape, just our URL and your key.

Quick Setup Guide

{
  "mcpServers": {
    "pmi": {
      "type": "http",
      "url": "https://pmi-api-beta-7tvvt.ondigitalocean.app/mcp",
      "headers": {
        "Authorization": "Bearer pmi_live_YOUR_KEY_HERE"
      }
    }
  }
}

Testing your connection

If a tool isn't behaving, run these cURL commands directly against the server to confirm everything's wired up correctly.

1. Health check (no auth)

The simplest “is the server alive?” check. No API key needed, plain JSON response.

curl https://pmi-api-beta-7tvvt.ondigitalocean.app/health

Expected response:

{"status":"ok"}

2. List available tools (free, requires API key)

Returns the full schema of all 28 tools. No credits used.

curl -X POST https://pmi-api-beta-7tvvt.ondigitalocean.app/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Authorization: Bearer pmi_live_YOUR_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

3. Call a tool (uses credits)

Example: get the rental yield for outcode LS6 (2 credits).

curl -X POST https://pmi-api-beta-7tvvt.ondigitalocean.app/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Authorization: Bearer pmi_live_YOUR_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"pmi_rents_yields","arguments":{"outcode":"LS6"}}}'

Technical reference

Server information

Location inputs

Most location-based tools accept input in one of these formats. Provide only ONE per request:

Building your own agent?

If you're rolling your own MCP client, the minimal connection sequence is initialize → tools/list → tools/call. Same Bearer token throughout.

Auth, billing & limits

• Auth: standard Authorization: Bearer <api_key> header. Same key as REST.
• Credits: deducted on 2xx tool responses. Same credit weights as the REST endpoints (see table above). All MCP usage shows up in GET /v1/account/usage under the existing endpoint names. There's no separate “MCP usage” view.
• Rate limit: per-10-second window, scoped to your API key. MCP and REST share the same bucket.
• Errors: MCP tool responses always come back as text (success or readable error). 401, 402, 429, 404 from the REST API become natural-language messages the LLM can quote.

FAQ

Troubleshooting