MCP
Connect an AI assistant to the Stream Estate API through the MCP server
Stream Estate exposes an MCP (Model Context Protocol) server at https://api-v2.stream.estate/mcp. It lets an AI assistant — Claude Code, Claude Desktop, Cursor… — query the API in natural language. The transport is HTTP and authentication uses the same X-API-KEY as the rest of the API.
Connection
claude mcp add --transport http streamestate https://api-v2.stream.estate/mcp \
--header "X-API-KEY: <your_key>"Authentication
The MCP server uses the same key as REST calls, passed in the X-API-KEY header. See Authentication to get and use it.
Treat your key like a password. Never commit it to a public Git repository, and don't paste it into a shared configuration.
What the assistant can do
Through the MCP server, the assistant has six tools. Each one goes through the matching REST route, with your key and your permissions:
| Tool | Job | Type |
|---|---|---|
search_properties | Search properties with a criteria tree | Read |
resolve_location | Resolve a place name or postal code to an administrative division | Read |
list_alerts | List the account's alerts | Read |
create_alert | Create an alert | Write |
update_alert | Update an existing alert, named fields only | Write |
delete_alert | Delete an alert | Write |
Every tool declares its behaviour per the MCP specification (title, readOnlyHint, destructiveHint, idempotentHint, openWorldHint). An assistant that honours these hints, Claude included, runs the read tools without asking for confirmation and asks before any write. The underlying routes are described in the API Reference.
Usage and billing
Calls made through MCP count against the same key and quota as direct calls. They show up in GET /account/api-usage under the byChannel breakdown, where channel is "MCP" (as opposed to "DIRECT"), with the number of calls and results per channel. See Usage tracking.