mcp
Distribution metadata and client examples for the Immopix MCP connector
Documentation
Immopix MCP connector
Immopix is an AI photo service for real estate listings. It improves listing photos: it replaces a
grey sky, removes clutter, adds furniture to an empty room and adjusts the season of a garden shot.
The customers are estate agents in Germany and, in English, elsewhere in Europe.
The Immopix connector is a remote MCP server at `https://immopix.ai/mcp`. It links an Immopix
account to Claude, ChatGPT and other assistants that support MCP connectors, so the agent can work
with the properties and photos that account already holds.
This repository holds the distribution metadata and the client examples for that connector. The
hosted server itself is closed source, and none of its code is published here.
What the connector can do
The server publishes thirteen tools. Grouped by what a user asks for, they cover:
- Properties. List the account's properties, search them by name, read one property, create a
property, rename it and delete it.
- Photos. List the photos of a property, read one photo with its edit history, and delete a
photo.
- Enhancement. Start an enhancement for a photo that is already in the account, then poll the
job until it returns a link to the finished image.
- Options and cost. Read the available presets, room types and furniture styles, the credit
price of the next enhancement, and the account's remaining credit balance.
- New photos. Return a link to the Immopix upload page for a property. Photos are uploaded in
the browser, never through the connector.
| Tool | Scope | Purpose |
|---|---|---|
| `immopix_list_properties` | read | List properties, newest first, with an optional name filter |
| `immopix_get_property` | read | Read one property |
| `immopix_list_property_photos` | read | List the photos of a property |
| `immopix_get_photo` | read | Read one photo, its revisions and the cost of the next enhancement |
| `immopix_check_job` | read | Read the status of an enhancement job and its result link |
| `immopix_list_presets` | read | List the presets, room types and furniture styles |
| `immopix_get_account` | read | Read the remaining credit balance |
| `immopix_get_upload_link` | write | Return the browser upload link for a property |
| `immopix_create_property` | write | Create a property |
| `immopix_update_property` | write | Rename a property |
| `immopix_delete_property` | write | Delete a property and its photos |
| `immopix_delete_photo` | write | Delete a photo and its revisions |
| `immopix_enhance_photo` | write | Enhance a photo in the account and return a job handle |
The full catalog stays visible with the `read` scope. Calling a write tool starts a step-up
authorization instead of failing silently.
Connecting
Setup happens inside the assistant and takes about a minute. The menu names differ between
assistants, the steps do not.
1. Open the connector settings of your assistant.
2. Add a custom connector with the address `https://immopix.ai/mcp`.
3. Sign in with your Immopix account in the browser window that opens.
4. The approval screen names the application, the account and the requested permissions. Check
them, then approve.
In Claude Code the same connector is added on the command line:
claude mcp add --transport http immopix https://immopix.ai/mcpRun `/mcp` afterwards to complete the sign-in.
`examples/README.md` walks through a first conversation and lists the client configurations.
Authentication
The server is an OAuth 2.1 protected resource. There is no API key and no client secret in this
flow.
- Clients register through Client ID Metadata Documents (CIMD) as public clients with
`token_endpoint_auth_method=none`. Dynamic client registration is not enabled.
- The authorization code flow requires PKCE with `S256`.
- The protected resource value is exactly `https://immopix.ai/mcp`, and it is bound to the issued
token. A token issued for another resource is rejected.
- The first grant asks for `read`. A write tool triggers one step-up for `read write`, so an agent
cannot change or charge anything under a read-only grant.
- Access tokens are valid for one hour. Refresh tokens rotate and are valid for 30 days.
- Discovery documents live at `/.well-known/oauth-protected-resource/mcp`,
`/.well-known/oauth-protected-resource` and `/.well-known/oauth-authorization-server`.
The endpoint accepts `POST` with `Authorization: Bearer` only. `https://immopix.ai/mcp/` with a
trailing slash is not the endpoint and is not redirected.
Photos
No image data travels over the MCP connection. The connector accepts neither an image URL nor
base64 bytes, and every enhancement works on a photo the account already holds.
To add new photos, ask the assistant for an upload link. `immopix_get_upload_link` returns the
Immopix page for that property, you upload the files in the browser, and the assistant lists the
photos again to pick up the new ones. Results come back as short-lived links, so the images stay
out of the model context.
Credits
An enhancement started through an assistant costs the same credits as the same enhancement in the
Immopix web interface. Reading, listing and searching cost nothing.
The tool descriptions and the server instructions require the assistant to name the price and the
balance and to ask for confirmation before it spends anything. `immopix_get_photo` returns the cost
of the next enhancement per output resolution, and a free revision honestly reports zero.
`immopix_get_account` returns the balance. Every enhancement also appears in the photo history in
Immopix.
Create and enhance calls carry a client-supplied `operation_id`. A repeated call with the same ID
and the same arguments replays the first result and charges nothing a second time.
Contents of this repository
| File | Purpose |
|---|---|
| `server.json` | Metadata for the official MCP Registry |
| `examples/` | A first conversation and client configurations |
| `CHANGELOG.md` | Published server and tool metadata versions |
| `SECURITY.md` | How to report a vulnerability |
| `PRIVACY.md` | Links to the binding privacy policy |
The hosted server is closed source. This repository carries distribution metadata and client
examples only. It contains no server code, no configuration and no credentials.
Links
- Help, German: https://immopix.ai/docs/mcp/
- Help, English: https://immopix.ai/en/docs/mcp/
- REST API documentation: https://immopix.ai/docs/api/
- Privacy, binding German version: https://immopix.ai/datenschutz/
- Privacy, English: https://immopix.ai/en/privacy/
- Terms, binding German version: https://immopix.ai/agb/
- Terms, English: https://immopix.ai/en/terms/
- Support: contact@immopix.ai
Frequently asked questions
What is mcp?
mcp is Distribution metadata and client examples for the Immopix MCP connector
How do I install mcp?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is mcp open source?
Yes — it is hosted on GitHub at https://github.com/Immopix/mcp.
Related MCP tools
🔌 Plug-and-play library to enable agents to call MCP and UTCP tools via code execution.
A MCP for Claude Desktop / Claude Code / Windsurf / Cursor to build n8n workflows for you
🚀 The fast, Pythonic way to build MCP servers and clients Trusted by 19900+ developers. Trusted by 19900+ developers. Trusted by 19900+ developers.
MCP server to provide Figma layout information to AI coding agents like Cursor
Mcp-use is the easiest way to interact with mcp servers with custom agents TypeScript-based implementation. Trusted by 8100+ developers.
Agent Framework For Fintech for the Model Context Protocol. Enhance AI assistants with powerful integrations. Python-based implementation.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP