cenogram-mcp-server
Official# Cenogram MCP Server
[](https://www.npmjs.com/package/@cenogram/mcp-server)
[](https://nodejs.org)
[](LICENSE)
**Polish Real Estate Transaction & Parcel Data for AI**
MCP server for Polish real estate data. Access 8M+ real estate transactions from the national Registry of Prices and Values (Rejestr Cen Nieruchomosci, RCN) - prices from notarial deeds, not listings - directly from Claude, Cursor, ChatGPT, Grok, or any MCP-compatible AI assistant. Beyond transaction prices, the server resolves cadastral parcels and adds per-parcel context: zoning, flood and landslide risk, heritage register, building permits and construction activity, public transport access, agricultural land classification, surrounding land use, nature - nearby forest and overlapping protected areas - what lies underground (mining terrains and major groundwater reservoirs), the land-use and soil-quality classification, and road access.
Data source: Polish national RCN registry (Rejestr Cen Nieruchomosci) | Platform: [cenogram.pl](https://cenogram.pl?src=mcpstdio)
## Get your API key
1. Go to [cenogram.pl/api](https://cenogram.pl/api?src=mcpstdio)
2. Enter your email
3. You'll receive your `cngrm_...` API key by email
Manage your keys at [cenogram.pl/ustawienia](https://cenogram.pl/ustawienia).
### Free tier
New accounts start with 1,000 tokens and a 14-day trial of the Standard plan. After that the key keeps
working on 50 tokens a week - the balance is restored to 50 every 7 days rather than accumulating -
with no expiry date and no card.
A call costs 1 token for statistics, reference data, parcel identity and municipal context; 3 for a
transaction search (the list plus its summary; 1 with `summaryOnly`); 2 for a price histogram or a
list of the parcels in an area; 4 for one context layer of
a parcel or transaction; 5 for spatial search, parcel outlines, valuation and a multi-district
comparison; 45 for `get_parcel_report`, which returns every layer at once. Location catalogues are
free. Paid plans raise the allowance - see [cenogram.pl/api](https://cenogram.pl/api?src=mcpstdio#cennik).
## Installation
Pick your client. All options below use the hosted server - no local install needed (except npx/stdio).
<details open>
<summary><strong>Claude Code</strong></summary>
One command - zero config files:
```bash
claude mcp add cenogram https://mcp.cenogram.pl/mcp \
-t http -H "Authorization: Bearer YOUR_API_KEY"
```
</details>
<details>
<summary><strong>Codex CLI</strong></summary>
**API key.** Add to `~/.codex/config.toml`:
```toml
[mcp_servers.cenogram]
url = "https://mcp.cenogram.pl/mcp"
bearer_token_env_var = "CENOGRAM_API_KEY"
```
Then export the key in the shell that starts Codex:
```bash
export CENOGRAM_API_KEY=cngrm_...
```
**OAuth.** Use the same block without `bearer_token_env_var`, then sign in:
```toml
[mcp_servers.cenogram]
url = "https://mcp.cenogram.pl/mcp"
```
```bash
codex mcp login cenogram
```
Without an API key or an OAuth login the server answers 401. Configuration keys differ between Codex versions, so check the [Codex documentation](https://developers.openai.com/codex/mcp) for your version; this section does not guarantee that a particular version connects.
</details>
<details>
<summary><strong>Cursor</strong></summary>
Add to `.cursor/mcp.json` in your project:
```json
{
"mcpServers": {
"cenogram": {
"type": "http",
"url": "https://mcp.cenogram.pl/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}
```
</details>
<details>
<summary><strong>Claude Desktop</strong></summary>
Add to your config file:
- macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`
- Windows: `%APPDATA%\Claude\claude_desktop_config.json`
**npx (stdio):**
```json
{
"mcpServers": {
"cenogram": {
"command": "npx",
"args": ["-y", "@cenogram/mcp-server@latest"],
"env": {
"CENOGRAM_API_KEY": "YOUR_API_KEY"
}
}
}
}
```
</details>
<details>
<summary><strong>VS Code / GitHub Copilot</strong></summary>
Add to `.vscode/mcp.json` in your workspace:
```json
{
"servers": {
"cenogram": {
"type": "http",
"url": "https://mcp.cenogram.pl/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}
```
</details>
<details>
<summary><strong>Windsurf</strong></summary>
Add to `~/.codeium/windsurf/mcp_config.json`:
**HTTP remote:**
```json
{
"mcpServers": {
"cenogram": {
"type": "http",
"url": "https://mcp.cenogram.pl/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}
```
If HTTP doesn't work, use the **npx (stdio)** option below instead.
</details>
<details>
<summary><strong>Cline</strong></summary>
In VS Code: Settings > Cline > MCP Servers. Add:
```json
{
"cenogram": {
"type": "http",
"url": "https://mcp.cenogram.pl/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
```
</details>
<details>
<summary><strong>npx (stdio) - local/offline</strong></summary>
Requires **Node.js >= 18**. Use this if you want to run the server locally instead of connecting to the hosted one.
```json
{
"mcpServers": {
"cenogram": {
"command": "npx",
"args": ["-y", "@cenogram/mcp-server@latest"],
"env": {
"CENOGRAM_API_KEY": "YOUR_API_KEY"
}
}
}
}
```
| Client | Config file |
|---|---|
| Cursor | `.cursor/mcp.json` |
| Claude Code | `.mcp.json` in your project |
| Claude Desktop | `~/Library/Application Support/Claude/claude_desktop_config.json` |
| Windsurf | `~/.codeium/windsurf/mcp_config.json` |
| Cline | VS Code settings > Cline > MCP Servers |
</details>
## Configuration
| Env Variable | Required | Default | Description |
|---|---|---|---|
| `CENOGRAM_API_KEY` | **Yes** (stdio) | - | API key from [cenogram.pl/api](https://cenogram.pl/api?src=mcpstdio) |
| `CENOGRAM_API_URL` | No | `https://cenogram.pl` | API base URL |
| `MCP_TRANSPORT` | No | `stdio` | Set to `http` for Streamable HTTP mode |
| `MCP_PORT` | No | `3002` | HTTP server port (HTTP mode only) |
| `CENOGRAM_CLIENT_ID` | No | auto-generated | Persistent client identifier |
You can also use the `--http` CLI flag instead of `MCP_TRANSPORT=http`.
## Tips
- **Model selection**: For best results, use Claude **Opus**. It makes more sequential tool calls and produces richer analysis. You can switch the model in the dropdown at the bottom of the chat window.
## Example Prompts
**Polish:**
- "Jaka jest mediana cen mieszkan w Krakowie w 2025?"
- "Pokaz transakcje z ulicy Pulawskiej 15 na Mokotowie"
- "Znajdz transakcje na dzialce 126104_9.0015.201"
- "Sprawdz plan miejscowy i ryzyko powodziowe dla dzialki 126104_9.0015.201"
- "Znajdz transakcje gruntow w promieniu 5km od centrum Wroclawia powyzej 500 000 PLN"
- "Porownaj ceny mieszkan na Mokotowie i Woli"
- "Pokaz rozklad cen nieruchomosci w Polsce"
**English:**
- "What's the median apartment price in Krakow in 2025?"
- "Show transactions at Pulawska 15 in Mokotow"
- "Find all transactions on parcel 126104_9.0015.201 and then search nearby"
- "Check the zoning and flood risk for parcel 126104_9.0015.201"
- "Find land transactions within 5km of Wroclaw center above 500,000 PLN"
- "Compare apartment prices in Mokotow and Wola districts"
- "Show the price distribution of real estate in Poland"
## Tools
| Tool | Description | Key Parameters |
|---|---|---|
| `search_transactions` | Search transactions with filters | location, street, buildingNumber, parcelId, propertyType, marketType, price/date/area range |
| `get_price_statistics` | Price/m2 stats by location (residential only); last 24 months by default, the answer states the period | location (optional), period (`24m`\|`all`), dateFrom/dateTo |
| `get_price_distribution` | Price histogram | bins, maxPrice |
| `search_by_area` | Search transactions by geographic radius | latitude, longitude, radiusKm |
| `get_market_overview` | Database overview and stats | (none) |
| `list_locations` | List available locations | search (optional) |
| `search_parcels` | Search parcels by cadastral ID prefix | q (parcel ID prefix, min 3 chars) |
| `list_parcels_in_area` | List the cadastral parcels in an area - light list or full outlines | teryt, location, bbox, lat + lng + radiusKm, or polygon; includeGeometry, minArea/maxArea, street/buildingNumber, cursor |
| `search_by_polygon` | Search transactions within a GeoJSON polygon | polygon, propertyType, dateFrom/dateTo |
| `compare_locations` | Compare stats across 2-5 districts | districts (comma-separated), propertyType |
| `get_building_breakdown` | Per-building breakdown for one transaction (footprint, storeys, est. floor area) | transaction_id (UUID from a search result) |
| `get_parcel_report` | Composite dossier for one parcel: core, 13 enrichment layers, transaction history, local price context (median zł/m² per property type for the county and the gmina), municipal context and a parcel-shape section (bounding rectangle, widest inscribed circle, shape class, area away from the boundaries - geometry of the outline only) | parcelId (cadastral id or UUID) |
| `resolve_parcel` | Resolve a parcel to its cadastral identity | parcelId, q (full cadastral id or 'locality + parcel number' - not a street address), or lat + lng |
| `get_demographics` | Population and demographic context for a location | location or teryt, year (or yearFrom/yearTo), category |
| `get_infrastructure_signals` | Municipal infrastructure signals (tenders, utilities, capital spending) | location or teryt |
| `estimate_value` | Comparable-sales value estimate for an apartment | area, plus lat + lng or parcelId; rooms, market |
| `estimate_land_value` | Comparable-sales value estimate for undeveloped land (building plot or farmland), per m² and per hectare | lat + lng (area required) or parcelId (area from the cadastre); landType, mpzpDesignation |
| `get_transaction_flood` | Flood risk for the property in a transaction | transaction_id (UUID from a search result) |
| `get_transaction_heritage` | Heritage-register status for the property | transaction_id |
| `get_transaction_landslide` | Landslide risk for the property | transaction_id |
| `get_transaction_surroundings` | Nuisance and land-use context around the property | transaction_id |
| `get_transaction_transit` | Public transport accessibility for the property | transaction_id |
| `get_transaction_permits` | Building permits recorded for the property | transaction_id |
| `get_permits_in_area` | Building permits and works notifications registered since 2016 around a point (issued decisions only), as a list plus a year × category aggregate | lat + lng; radiusKm (max 2), dateFrom, dateTo, intent, categories, excludeCategories, worksType, limit, cursor, aggregateOnly |
| `get_transaction_planning` | Local zoning and planning status for the property | transaction_id |
| `get_transaction_farmland` | Agricultural land-use classification for the property | transaction_id |
| `get_transaction_nature` | Nearby forest and overlapping protected natural areas for the property | transaction_id |
| `get_transaction_subsurface` | Mining terrains and major groundwater reservoirs under the property | transaction_id |
| `get_transaction_roads` | Geometric evidence of road access for the property (distances, road class, access indicator) | transaction_id |
### Location naming
- Most cities: use the city name directly (e.g., "Gdansk", "Lublin")
- Warsaw: "Warszawa" covers all 18 districts at once; name one ("Mokotow", "Srodmiescie", "Wola") to narrow it down
- Krakow and Lodz work the same way: the city name covers every sub-district, or name one ("Krakow-Podgorze")
- Neighbourhood names are not administrative units - search by radius or polygon instead
- Use `list_locations` to find valid names
### Property types
| Value | Polish | English |
|---|---|---|
| `land` | Grunt | Land plot |
| `building` | Budynek | Building |
| `developed_land` | Grunt zabudowany | Developed land |
| `unit` | Lokal | Apartment/unit |
### Workflows
Results include parcel IDs and GPS coordinates, enabling multi-step research:
```
1. Search by address -> search_transactions(location="Mokotow", street="Pulawska", buildingNumber="15")
2. Note parcel_id and coordinates from results
3. Search nearby -> search_by_area(lat=52.19, lng=21.01, radiusKm=2, propertyType="unit")
4. Compare prices -> get_price_statistics(location="Mokotow")
```
This mimics how a property appraiser finds comparable transactions for valuation reports.
Starting from the land rather than from a deed is a different route:
```
1. List the parcels -> list_parcels_in_area(location="Wawer", minArea=800)
2. Draw the ones you want -> list_parcels_in_area(bbox="21.10,52.20,21.14,52.23", includeGeometry=true)
3. Open one in full -> get_parcel_report(parcelId="146518_8.0108.27")
```
Step 1 pages by cursor; step 2 is capped and usually truncated, so narrow the box rather than reading
a truncated answer as the parcel list of the area.
## Data
- **8M+ transactions** from all of Poland (380 counties)
- **Date range:** 2003 - present
- **Source:** Polish national RCN registry (Rejestr Cen Nieruchomosci)
- **Refresh:** periodic updates from RCN
- **Per-parcel context:** zoning, flood and landslide risk, heritage register, building permits and construction activity, transit access, agricultural land use, surroundings, nature (nearby forest and protected areas), subsurface (mining terrains and major groundwater reservoirs), land-use and soil-quality classification, and road access - addressable by cadastral ID
## Troubleshooting
**"Error: CENOGRAM_API_KEY is required"** - This only applies to stdio mode. Make sure `CENOGRAM_API_KEY` is set in the `env` block of your MCP config. For HTTP remote, the key goes in the `Authorization` header instead.
**npx hangs or fails** - Check your Node.js version with `node -v`. The stdio mode requires Node.js >= 18. If you're on an older version, use the HTTP remote option instead (no Node.js needed).
**A location returns 0 results** - The name may not be an administrative unit. Districts and neighbourhoods are two different things: "Mokotow" is a district and works, "Sluzew" is a neighbourhood inside it and does not. Use `list_locations(search="...")` to find valid names, or search by radius (`search_by_area`) for anything smaller than a district.
**401 Unauthorized (HTTP mode)** - The server accepts either an API key (`Authorization: Bearer cngrm_...`, with the `Bearer` prefix and the full key, not just the prefix) or an OAuth access token from a client that completed the OAuth sign-in. A missing token, or one in an unrecognised format, gets a 401 whose body links to [cenogram.pl/api](https://cenogram.pl/api?src=mcpstdio), where you get a key. A client with neither a key nor an OAuth login (Codex CLI, for example) gets 401 on every request - see its setup above.
**402 Payment Required** - The account is out of tokens. On a free account the response says the date from which the balance is back at its weekly 50; the key and the connection stay valid until then. A paid plan lifts the limit - see [cenogram.pl/api](https://cenogram.pl/api?src=mcpstdio#cennik).
## Development
```bash
git clone https://github.com/cenogram/mcp-server.git
cd mcp-server
npm install
npm test
npm run build
```
## License
MIT
TDQS
Scored across 30 tools
Descriptions actively steer callers between overlapping families (search_transactions vs search_by_area vs search_by_polygon; get_parcel_report vs per-layer get_transaction_* tools), so most boundaries are clear. However, with 30 tools and several near-neighbors (get_transaction_permits vs get_permits_in_area, estimate_value vs estimate_land_value), an agent still needs to read carefully to avoid misselection.
Nearly all tools use a predictable snake_case verb_noun pattern (search_transactions, get_price_statistics, get_transaction_flood, list_locations). The main deviations are the search_by_area/search_by_polygon forms and mixed estimate_* naming, but these remain readable and not chaotic.
The domain is broad, but 30 tools exceeds the recommended range and includes many per-layer transaction detail tools that largely duplicate what get_parcel_report returns. Several could be consolidated into a parameterized transaction-detail or parcel-report call.
The read-only surface is extensive: transaction search by place/radius/polygon, price statistics, parcel resolution and reporting, environmental layers, permits, planning, nature, farmland, valuation, demographics and infrastructure. A generic get_transaction(id) endpoint is notable by its absence, but the layer-specific and aggregate report tools cover most workflows without major dead ends.