MCP — for Claude and developer tools

LetsRealty runs a remote Model Context Protocol server, so an AI assistant can work your CRM in conversation: "which Hot Leads haven't we called this week?", "log this WhatsApp chat as a note on Ana's Lead".

  • With an API Key: https://api.letsrealty.io/mcp (Streamable HTTP), with Authorization: Bearer lr_live_…
  • Signed in as a User (OAuth): https://<your-slug>.admin.letsrealty.io/mcp, your Admin Console's address. Each User finds theirs in Settings › AI connections.

Both follow the MCP specification up to version 2026-07-28. Calls through MCP cost LetsRealty no AI and do not count toward your AI allowance or the Assistant's monthly tasks. They do count toward your rate limits.

Sign in as a User (OAuth)

Claude and ChatGPT connectors sign in with OAuth 2.1, as the MCP authorization spec describes. Add your console's /mcp URL as a custom connector, and the app does the rest: a 401 from /mcp points it at the Protected Resource Metadata (RFC 9728), which names your console as the authorization server (RFC 8414 metadata at /.well-known/oauth-authorization-server). The app identifies itself with a Client ID Metadata Document (an https client_id), or registers with Dynamic Client Registration (/oauth/register). The User signs in on your console and allows it on a consent screen. Authorization uses PKCE (S256) and an exact redirect_uri, and tokens are bound to that console's /mcp (RFC 8707 resource): they work nowhere else. Refresh tokens rotate on every use, and a reused one ends the connection.

A signed-in app gets the tools of the in-console Assistant for that User's role, with exactly that User's permissions: an Agent changes only their own listings and never sees the website tools. Every tool sets all four hints; every tool that changes something is marked destructive, so the app asks before running it. A change runs directly, comes back with links to what changed, and is logged as "via <app>". The User can undo it from Settings › AI connections for 24 hours, and disconnect the app there at any time.

Supported clients: Claude (every plan) and ChatGPT Business and Enterprise. Use an API Key for automations such as Zapier.

API Keys: choose the key's scopes

The tools a key can use follow its scopes. For an assistant that only answers questions, give it leads:read, activities:read and properties:read. Add leads:write, activities:write and properties:write only if it should change things — tools that change status, archive or reassign a Lead, or archive a listing, are marked as destructive, so your client asks you before running them.

Tools

  • search_leads (leads:read) — Find Leads by status, Lead Score tier, owner, updated_since and text.
  • get_lead (leads:read) — One Lead with its recent Inquiries and, with activities:read, its recent Activities.
  • list_inquiries (leads:read) — Inquiries, optionally of one Lead or about one Property.
  • list_open_activities (activities:read) — Activities not yet completed, soonest due first; overdue, today or week.
  • ingest_lead (leads:ingest) — Record that someone got in touch, as POST /v1/leads:ingest does.
  • add_note (activities:write) — Log a note on a Lead.
  • create_task (activities:write) — Schedule a task on a Lead.
  • complete_activity (activities:write) — Mark an Activity done.
  • update_lead_status (leads:write) — Move a Lead to another stage (destructive); to closed_lost it needs a lost_reason, and other a lost_reason_note.
  • archive_lead (leads:write) — Archive a Lead (destructive).
  • assign_owner (leads:write) — Set or clear the Lead Owner (destructive).
  • pin_lead_tier (leads:write) — Pin a Lead's tier by hand (Hot, Warm, Moderate or Cold); Lead Scoring keeps scoring underneath and never changes it.
  • clear_lead_tier (leads:write) — Clear a Lead's pinned tier, so its tier is Lead Scoring's again.
  • search_properties (properties:read) — Find listings by text, listing_type, place_code and updated_since, archived ones too with include_archived, in a Site Language with lang.
  • get_property (properties:read) — One listing, in a Site Language with lang.
  • find_matching_leads (leads:read) — The active Leads whose saved criteria a listing meets.
  • create_property (properties:write) — Create a listing, as POST /v1/properties does.
  • update_property (properties:write) — Change a listing's fields.
  • add_property_image_from_url (properties:write) — Add a photo to a listing from an https URL, as POST /v1/properties/{property_id}/images:import does.
  • archive_property (properties:write) — Archive a listing (destructive).
  • market_answer (market:read) — What listings on the market ask for a Place, Property Type, beds and baths, as GET /v1/market/answer does; place takes a Place code or a canton or district name. See Market answers.

A key is only shown the tools its scopes allow. Results use the same shapes as the API reference. Each MCP request counts against your rate limits like any API call, ingest_lead also uses the ingestion budget, and market_answer the market answers' budget. ingest_lead, add_note, create_task, create_property and update_property take an optional idempotency_key: a retry with the same key and arguments returns the first result instead of writing twice.

Claude Code

claude mcp add --transport http letsrealty https://api.letsrealty.io/mcp \
  --header "Authorization: Bearer $LETSREALTY_API_KEY"

Claude Desktop, Claude.ai and ChatGPT

Add a custom connector with your console's URL, https://<your-slug>.admin.letsrealty.io/mcp, and sign in when it asks, as described above. No key is needed.

Cursor and other clients

Any client that accepts static headers works. For Cursor, in .cursor/mcp.json:

{
  "mcpServers": {
    "letsrealty": {
      "url": "https://api.letsrealty.io/mcp",
      "headers": { "Authorization": "Bearer lr_live_…" }
    }
  }
}

Never available

Browsing history and website chat transcripts are never available over MCP or the API.