mcp-csu
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mcp-csu Search datasets for inflation rate"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mcp-csu
MCP server for the Czech Statistical Office (ČSÚ / CZSO) DataStat API. Gives AI assistants direct access to 700+ statistical datasets about the Czech Republic — population, economy, prices, wages, employment, industry, agriculture, trade, tourism, environment, and more.
Single Python file. No cloning required — just uvx mcp-csu.
Features
Full catalog access — search, browse, and inspect all 700+ datasets and 1500+ predefined tables
Data retrieval — fetch statistical data as CSV, query individual values with full context
AI-optimized output — human-readable text for metadata, CSV for data, automatic truncation with row counts
Rate limiting — built-in concurrency control (3 parallel requests) and minimum request interval (150ms)
Caching — catalog listings cached in memory for 10 minutes to avoid redundant requests
No authentication — the DataStat API is public
Related MCP server: datacube-sk
Prerequisites
uv (Python package runner)
That's it. Python and all dependencies are managed automatically by uv.
Configuration
Claude Code
Add to ~/.claude/settings.json:
{
"mcpServers": {
"csu": {
"command": "uvx",
"args": ["mcp-csu"]
}
}
}Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"csu": {
"command": "uvx",
"args": ["mcp-csu"]
}
}
}Any MCP client
The server uses stdio transport (default). Launch command:
uvx mcp-csuData model
The DataStat database has a hierarchical structure:
Dataset (sada) e.g. CEN0101H — "Míra inflace"
├── Dimensions (dimenze) e.g. CasR (years), Uz0 (territory)
│ └── Items (položky) e.g. "2024", "CZ"
├── Indicators (ukazatele) e.g. 6134J06 — "Průměrná roční míra inflace"
└── Selections (výběry) e.g. CEN0101HT01 — "Průměrná roční míra inflace"
└── CSV data pre-configured table ready to fetchDatasets contain raw multidimensional data. Each dataset has dimensions (time, territory, categories) and indicators (what is measured).
Selections are predefined views — a specific slice of a dataset with fixed dimension arrangement. They are the easiest way to get data.
Tools
Discovery
search_datasets
Full-text search across all datasets. Returns dataset codes, names, time period types, and territory levels.
Parameter | Type | Required | Description |
| string | yes | Search keyword (Czech recommended) |
search_datasets("inflace")
→ Found 3 dataset(s):
WCEN01 (v4) — Index spotřebitelských cen (indexy, míra inflace)
WCEN01M (v1) — Index spotřebitelských cen — měsíční data
CEN0101H (v1) — Míra inflacesearch_selections
Full-text search across all predefined data tables.
Parameter | Type | Required | Description |
| string | yes | Search keyword (Czech recommended) |
search_selections("mzdy")
→ Found 30 selection(s):
MZDQ1T1 — Průměrný evidenční počet zaměstnanců a průměrné hrubé měsíční mzdy...
Period: Čtvrtletí | Territory: Stát | Dataset: MZDQ1list_datasets
Paginated listing of all datasets.
Parameter | Type | Default | Description |
| int | 0 | Skip first N items |
| int | 30 | Items per page (max 100) |
list_selections
Paginated listing of all predefined tables.
Parameter | Type | Default | Description |
| int | 0 | Skip first N items |
| int | 30 | Items per page (max 100) |
Exploration
get_dataset
Full dataset metadata: dimensions with item counts, indicators with definitions, keywords, update frequency.
Parameter | Type | Required | Description |
| string | yes | Dataset code (e.g. |
get_dataset("CEN0101H")
→ Dataset: CEN0101H (v1)
Name: Míra inflace
Keywords: míra inflace
Update frequency: MONTHLY
Dimensions (4):
CasM — Měsíce (720 items)
CasR — Roky (61 items)
CASRMX — Měsíce, roky (780 items)
Uz0 — Území (1 items)
Indicators (4):
6134J09 — Přírůstek průměrného ročního indexu spotřebitelských cen - měsíční
6134J06 — Průměrná roční míra inflace
...get_dataset_selections
List predefined data tables for a specific dataset.
Parameter | Type | Required | Description |
| string | yes | Dataset code |
get_dataset_selections("CEN0101H")
→ Selections for CEN0101H (2):
CEN0101HT01 — Průměrná roční míra inflace
Period: Rok | Territory: Stát
CEN0101HT02 — Míra inflace - měsíční
Period: Měsíc | Territory: Státget_dimension_items
Get all possible values for a dimension. Supports hierarchy level filtering and pagination.
Parameter | Type | Default | Description |
| string | — | Dimension code from |
| string | null | Filter by hierarchy level (e.g. |
| int | 0 | Skip first N items |
| int | 50 | Items per page (max 200) |
get_dimension_items("UZ023H2U", level="KRAJ")
→ Dimension UZ023H2U — 14 item(s) at level KRAJ:
CZ010 — Hlavní město Praha (Capital City Prague) [KRAJ]
CZ020 — Středočeský kraj (Central Bohemian Region) [KRAJ]
CZ031 — Jihočeský kraj (South Bohemian Region) [KRAJ]
...get_indicator
Indicator definition and display format.
Parameter | Type | Required | Description |
| string | yes | Indicator code from |
Data retrieval
get_selection_data
Primary data access tool. Fetches CSV data from a predefined selection.
Parameter | Type | Default | Description |
| string | — | Selection code (e.g. |
| int | 100 | Max data rows. 0 = unlimited |
get_selection_data("CEN0101HT01", max_rows=5)
→ "Ukazatel","Území","Roky","Hodnota"
"Průměrná roční míra inflace","Česko","2025","2.5"
"Průměrná roční míra inflace","Česko","2024","2.4"
"Průměrná roční míra inflace","Česko","2023","10.7"
"Průměrná roční míra inflace","Česko","2022","15.1"
"Průměrná roční míra inflace","Česko","2021","3.8"
[Showing 5 of 29 rows. Use max_rows=0 for all data or max_rows=10 to see more.]get_value
Retrieve a single specific value. The most precise query — returns one number with full context (indicator name, dimension labels, publication date).
Parameter | Type | Default | Description |
| string | — | Dataset code |
| string | — | Indicator code |
| list[str] | — | Dimension codes in order |
| list[str] | — | Item codes matching dimensions |
| string | null | Dataset version (latest if omitted) |
get_value("RSO01", "3971b",
["CasR", "TYPPROSJED", "UZ023H2U"],
["2023", "501", "CZ"])
→ Value: 6 258
Indicator: Počet územních jednotek
Roky: 2023
Typ prostorové jednotky: Obec
ČR, kraje, okresy: Česko
Published: 2024-04-30T07:00:00Zcustom_query
Execute an arbitrary data query on a dataset. Returns CSV.
This is an advanced tool — prefer get_selection_data() when a suitable predefined table exists. The custom query API is sensitive to correct dimension placement and hierarchy level filtering.
Parameter | Type | Default | Description |
| string | — | Dataset code |
| string | — | Version from |
| list[dict] | — | Column dimensions (each needs |
| list[dict] | — | Row dimensions |
| list[dict] | null | Filter dimensions |
| int | 100 | Max CSV rows |
get_dataset_metadata
Dataset content statistics: record count, time range, publication and update timestamps.
Parameter | Type | Required | Description |
| string | yes | Dataset code |
| string | yes | Version from |
Usage examples
Get Czech inflation rate
1. search_datasets("inflace")
→ CEN0101H — Míra inflace
2. get_dataset_selections("CEN0101H")
→ CEN0101HT01 — Průměrná roční míra inflace
3. get_selection_data("CEN0101HT01")
→ CSV with annual inflation rates from 1994 to presentFind average wages by region
1. search_selections("mzdy kraje")
→ MZDQ1T2 — ... dle krajů a regionů soudržnosti
2. get_selection_data("MZDQ1T2", max_rows=20)
→ CSV with wages by regionGet exact population of Prague in 2023
1. search_datasets("obyvatelstvo")
→ OBY01 — Obyvatelstvo podle pohlaví a věku
2. get_dataset("OBY01")
→ see dimensions and indicators
3. get_dimension_items("<territory_dim>", level="KRAJ")
→ find Prague code
4. get_value("OBY01", "<indicator>",
["<time_dim>", "<territory_dim>"],
["2023", "<prague_code>"])
→ exact valueCzech vocabulary for search
The database is in Czech. Common search terms:
Czech | English | Example datasets |
obyvatelstvo | population | OBY01, OBY02 |
mzdy | wages | MZDQ1, MZD01 |
ceny | prices | CEN01, CEN02 |
inflace | inflation | CEN0101H |
HDP | GDP | NUC06R, NUC06Q |
nezaměstnanost | unemployment | ZAM04 |
průmysl | industry | PRU01 |
stavebnictví | construction | STA01 |
vzdělání | education | VZD01 |
zdraví | health | ZDR01 |
zemědělství | agriculture | ZEM01 |
doprava | transport | DOP01 |
cestovní ruch | tourism | CRU01 |
životní prostředí | environment | ZPR01 |
kriminalita | crime | KRI01 |
volby | elections | VOL01 |
bytová výstavba | housing | BYT01 |
zahraniční obchod | foreign trade | VZO01 |
Technical details
Architecture
Single-file Python server using FastMCP framework over stdio transport. Dependencies managed via PEP 723 inline script metadata — uv run installs them automatically into an isolated environment.
Upstream API
The server wraps two DataStat REST APIs:
API | Base URL | Purpose |
Catalog |
| Dataset/selection/dimension/indicator metadata |
Data |
| Data retrieval (CSV, JSON-STAT) |
API documentation:
Rate limiting
The DataStat API does not document rate limits, but the server applies conservative throttling:
Max concurrent requests: 3 (semaphore)
Min request interval: 150ms (global)
Request timeout: 60 seconds
Caching
Catalog listings (list_datasets, list_selections) are cached in memory with a 10-minute TTL. These endpoints return the full catalog (700–1500 items) on every call since the API ignores pagination parameters — caching avoids repeated large transfers.
Output formatting
Metadata tools return structured text with clear labels
Data tools return CSV (most compact and LLM-friendly tabular format)
Truncation: data responses are limited to 100 rows by default, with total count shown. Adjustable via
max_rowsparameterLanguage: all API responses are in Czech (
Accept-Language: cs)
Dependencies
Package | Version | Purpose |
| >=1.0.0 | MCP server framework (FastMCP) |
| >=0.27.0 | Async HTTP client |
Both installed automatically by uv run.
License
MIT
Available Tools
12 toolscustom_queryA
Execute a custom data query on a dataset (advanced).
IMPORTANT: Prefer get_selection_data() for predefined tables — it is much simpler and more reliable. Use custom_query only when no suitable predefined selection exists.
Caveats:
All dataset dimensions must be placed in columns, rows, or table_filters.
Datasets with multiple time dimensions (e.g., CasM + CasR + CASRMX) may only work with certain dimension combinations matching predefined selections.
Hierarchical territory dimensions may need "filtr" with "urovenHierarchieKod".
Args: dataset_code: Dataset code. dataset_version: Dataset version string from get_dataset(). columns: Column dimensions. Each dict must have "kodDimenze" (str). Optionally add "filtr" with [{"zobrazitPolozky": ["code1","code2"]}]. rows: Row dimensions. Same structure as columns. Use "kodDimenze": "#UKAZATEL" to put indicators as rows. table_filters: Header/filter dimensions. Same structure, but can also include "filtrTabulkyKod" (str) to filter to a single item. max_rows: Max CSV rows to return.
Inputs: dataset_code (required), dataset_version (required), columns (required), rows (required), +2 more. Returns: result
get_datasetA
Get detailed info about a dataset: description, dimensions, indicators, keywords.
Use this to understand dataset structure before querying data. The dimension codes and indicator codes are needed for get_value() and custom_query().
Inputs: dataset_code (required). Returns: result
get_dataset_metadataA
Get metadata about dataset content: record count, time range, last update.
Args: dataset_code: Dataset code. version: Dataset version from get_dataset().
Inputs: dataset_code (required), version (required). Returns: result
get_dataset_selectionsB
List predefined data tables for a specific dataset.
These selections can be fetched directly with get_selection_data(code). This is the recommended way to find available pre-built data views.
Inputs: dataset_code (required). Returns: result
get_dimension_itemsA
Get possible values for a dimension (e.g., years, regions, categories).
Args: dimension_code: Dimension code from get_dataset() output (e.g., CasR, Uz0). level: Filter by hierarchy level code (e.g., STAT, KRAJ, OKRES). offset: Skip first N items. limit: Max items to return (default 50, max 200).
Item codes are needed for get_value() and custom_query().
Inputs: dimension_code (required), level, offset, limit. Returns: result
get_indicatorC
Get detailed information about a statistical indicator.
Returns the indicator's full definition, display format, and related datasets.
Inputs: indicator_code (required). Returns: result
get_selection_dataA
Fetch actual statistical data from a predefined selection as CSV.
This is the primary and most reliable way to get data. Find selection codes via search_selections() or get_dataset_selections().
Args: selection_code: Selection code (e.g., CEN0101HT01). max_rows: Maximum number of data rows to return (default 100). Set to 0 for unlimited (use with caution — some tables are very large).
Inputs: selection_code (required), max_rows. Returns: result
get_valueA
Get a single specific value from a dataset.
This is the most precise way to query data — returns exactly one value. Requires knowing the exact dimension and item codes from get_dataset() and get_dimension_items().
Args: dataset_code: Dataset code (e.g., RSO01). indicator_code: Indicator code (e.g., 3971b). dimension_codes: List of dimension codes (e.g., ["CasR", "TYPPROSJED", "UZ023H2U"]). item_codes: List of item codes matching dimension_codes order (e.g., ["2023", "501", "CZ"]). version: Dataset version (optional, defaults to latest).
Example: Number of municipalities in Czech Republic in 2023: get_value("RSO01", "3971b", ["CasR","TYPPROSJED","UZ023H2U"], ["2023","501","CZ"])
Inputs: dataset_code (required), indicator_code (required), dimension_codes (required), item_codes (required), +1 more. Returns: result
list_datasetsA
List all available datasets with pagination.
Use offset and limit to page through results. Total: ~730 datasets.
Inputs: offset, limit. Returns: result
list_selectionsA
List all predefined data tables (selections) with pagination.
Selections are pre-configured data views that can be fetched directly. Use get_selection_data(code) to retrieve their data.
Inputs: offset, limit. Returns: result
search_datasetsA
Search for statistical datasets by keyword.
Search works best with Czech terms: obyvatelstvo=population, mzdy=wages, ceny=prices, nezaměstnanost=unemployment, průmysl=industry, HDP=GDP, inflace=inflation, vzdělání=education, zdraví=health.
Returns dataset codes usable with get_dataset() and get_dataset_selections().
Inputs: query (required). Returns: result
search_selectionsA
Search for predefined data tables (selections) by keyword.
Selections are pre-configured views of datasets — the easiest way to get data. Retrieve their data with get_selection_data(code). Search in Czech for best results.
Inputs: query (required). Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 12 tools
Each tool targets a distinct aspect of the CSU data workflow: discovery, metadata, dimension exploration, and data retrieval. Even the three query tools (get_value, get_selection_data, custom_query) are clearly differentiated by use-case guidance, making selection unambiguous.
All tools follow a consistent verb_noun pattern using snake_case (e.g., get_dataset, list_datasets, search_selections). The one outlier, custom_query, still adheres to the same pattern, so naming is uniform and predictable.
With 12 tools covering search, listing, metadata, dimension exploration, and multiple query methods, the set is well-scoped for the CSU domain. No tools feel redundant or missing for core operations.
The tools provide end-to-end coverage: dataset discovery (list/search), structure exploration (get_dataset, get_dimension_items, get_indicator), and data retrieval (predefined selections, custom queries, single values). No obvious gaps exist in the statistical data access lifecycle.
Maintenance
Related MCP Connectors
Official statistics for AI agents: Eurostat, World Bank, OECD, IMF and WHO data for 150+ countries.
Access European statistics: GDP, unemployment, inflation, population data.
Czech Statistical Office (Český statistický úřad, ČSÚ) open-data MCP.
Polish public statistics (GUS, NBP, Eurostat) with unit, period and source. Data tools only read.
1
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables querying Czech Statistical Office open data through MCP tools and natural language questions.95 npm2MIT
- AlicenseNot gradedqualityBmaintenanceEnables querying statistical data from the Statistical Office of the Slovak Republic via natural language or direct tool calls.85 npmMIT
- FlicenseNot gradedqualityBmaintenanceEnables AI agents to look up Czech companies, VAT status, entity KYC snapshots, address standardization, and Czech identifier validation via official ARES data.-
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to query official Statistics Denmark data without an API key — pulling national population counts and trends, browsing the subject tree, searching available tables, inspecting variable metadata, and retrieving detailed breakdowns by region, age, sex, and period in JSON-stat, bulk, or CSV format.84 npmMIT