Skip to main content
Glama
Thailand NSO MCP Server

Thailand NSO MCP Server

by aard-ai
WARNING

Research repository — not for data access.

aard-ai/tnso-mcp-server is a research artefact, built for studying how MCP servers behave against different statistical agencies' SDMX endpoints. It is not maintained, is not published to any package registry, and must not be used as an MCP server for data access.

Thailand NSO MCP Server

An MCP server that lets an LLM discover, query and analyse Thailand's official statistics from the National Statistical Office (NSO, agency TNSO) via the SDMX REST API — in natural language.

This is a lightweight MCP that uses keyword search (this is not the Aard MCP).
It should be effective for e.g. searching for indicators or dataflows.

Note that this lightweight approach is much less effective for certain question types such as "Compare Bangkok and Chon buri", or "Tell me about trade of all fish that are not tuna". These kinds of questions benefit from structure-based search across all data sources, just two factors that make the Aard MCP different from other tools.

When connected to your chat tool (e.g. Gemini or Claude) it can generate analysis like the below:

This implementation based on a port of ondata/istat_mcp_server (Italy / ISTAT), adjusted to point to the Thaliand NSO SDMX endpoint at https://ns1-stathub.nso.go.th/rest.

It uses a 9-tool workflow (like the ISTAT MCP), same two-layer caching; the data source, agency, languages (Thai/English) and geography (Thai provinces) are swapped in.

Note we use Buddhist Era dates. TNSO publishes time periods in the Buddhist Era calendar (BE = Gregorian + 543). So 2567 means 2024. Pass start_period / end_period as BE years.

Tools

Tool

What it does

discover_dataflows

Search ~900 TNSO dataflows by keyword (Thai/English; any keyword matches (OR), or match_all for AND), or by covers — the dimension codes a dataflow must actually have data for (e.g. {"CWT": ["10","20"]} → every dataflow carrying both Bangkok and Chon Buri).

get_structure

Dimensions + codelists for a data structure (DSD).

get_constraints

Valid values (with labels) per dimension + available time range. Start here.

get_codelist_description

Thai/English labels for every code in a codelist.

get_concepts

Resolve an SDMX concept id to its Thai/English name.

get_data

Fetch observations as a TSV table (+ reproducible CSV/curl URLs). On an empty result it self-diagnoses (invalid codes / out-of-range period) and suggests verified non-empty alternatives.

check_data_availability

Pre-flight check that a specific filter/period combination returns rows before a full get_data.

get_territorial_codes

Thai geography codes: region (CL_AREA), province (CL_CWT, 77 changwat), district (CL_AMPHOE).

get_cache_diagnostics

Inspect the on-disk cache.

Typical workflow: discover_dataflows → get_constraints → get_data (use get_territorial_codes first when you need province/region codes). get_data self-diagnoses empty results and suggests working alternatives; use check_data_availability to pre-check a combination before fetching.

Related MCP server: mcp-uk-ons

Install & run

Requires Python ≥ 3.11. Using uv (recommended):

git clone https://github.com/aard-ai/tnso-mcp-server.git
cd tnso-mcp-server
uv venv
uv pip install -e ".[dev]"

# Run the server (stdio transport)
uv run python -m tnso_mcp_server

Or with pip:

cd tnso-mcp-server
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
python -m tnso_mcp_server

Register with an MCP client

Claude Code:

claude mcp add tnso -- uv --directory /abs/path/to/tnso-mcp-server run python -m tnso_mcp_server

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "tnso": {
      "command": "uv",
      "args": ["--directory", "/abs/path/to/tnso-mcp-server", "run", "python", "-m", "tnso_mcp_server"]
    }
  }
}

Configuration

Copy .env.example to .env to override defaults (API base URL, timeouts, cache TTLs, log level, dataflow blacklist). All values have sensible defaults, so .env is optional.

Example

discover_dataflows(keywords="aging")
  -> DF_01DI_IND_AGING — "Aging Index dataflow"
discover_dataflows(covers={"CWT": ["10", "20"]})
  -> every dataflow whose data covers both Bangkok (10) and Chon Buri (20)
get_constraints(dataflow_id="DF_01DI_IND_AGING")
  -> dimensions POP_IND, SEX, AREA, CWT, ...; TIME_PERIOD range 2557–2567 (BE)
get_territorial_codes(level="province", name="bangkok")
  -> { code: "10", name_en: "Krung Thep Maha Nakhon (Bangkok)", name_th: "กรุงเทพมหานคร" }
get_data(dataflow_id="DF_01DI_IND_AGING", dimension_filters={"CWT": ["10"]}, start_period="2560", end_period="2567")
  -> TSV table of the aging index for Bangkok, 2017–2024

Tests

uv run pytest -m "not integration"   # fast unit tests (no network)
uv run pytest -m integration         # live tests against the real TNSO API
uv run pytest                         # everything

MIT (same as the upstream istat_mcp_server).

Available Tools

8 tools
discover_dataflowsA

Discover available TNSO (Thailand National Statistical Office) dataflows. Optionally filter by comma-separated keywords (matched against id, Thai/English name and description). Start here to find a dataset, then call get_constraints.

Inputs: keywords.

get_cache_diagnosticsA

Check persistent-cache health (exists, size, writability). Pass debug=true to additionally reveal the cache path and stored keys (host-sensitive). For debugging.

Inputs: debug.

get_codelist_descriptionA

Get Thai and English labels for every code in a codelist (e.g. 'CL_CWT').

Inputs: codelist_id (required).

get_conceptsA

Resolve a TNSO SDMX concept id to its Thai or English name (searches all TNSO concept schemes; result cached). Use lang='th' or lang='en'.

Inputs: concept_id (required), lang.

get_constraintsA

Get all valid dimension values (with Thai/English labels) and the available time range for a dataflow. One call returns everything needed to build a get_data query. Time periods are Buddhist Era (BE = Gregorian + 543).

Inputs: dataflow_id (required).

get_dataA

Fetch observations for a dataflow as a TSV table, with reproducible CSV/curl URLs. Supports dimension filtering and a time range. Omit dimensions from dimension_filters to get all their values; pass multiple codes per dimension as an array. If no period is given, the latest available year is returned. IMPORTANT: start_period/end_period are Buddhist Era years (e.g. 2567 = 2024).

Inputs: dataflow_id (required), detail, end_period, start_period, +2 more.

get_structureA

Get the data structure (DSD) for a datastructure id: the ordered list of dimensions and the codelist each uses. Fast path when you already know the codes.

Inputs: datastructure_id (required).

get_territorial_codesA

Look up Thai geography codes used by the AREA / CWT dimensions: level='region' (CL_AREA), 'province' (CL_CWT, 77 changwat), or 'district' (CL_AMPHOE). Filter by 'name' (Thai or English substring). Use this to find codes before calling get_data.

Inputs: name, level.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 8 tools

    Disambiguation5/5

    Each tool has a distinct purpose: discovery, structure, constraints, data retrieval, code lookups, and diagnostics. No overlap exists; even get_codelist_description and get_constraints serve different needs.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern with snake_case (e.g., discover_dataflows, get_constraints, get_data), making it predictable for agents.

    Tool Count5/5

    With 8 tools, the set is well-scoped for a statistical data server, covering discovery, structure, query building, data retrieval, and code lookups without unnecessary tools.

    Completeness5/5

    The tool surface covers the full workflow: discover dataflows, get structure and constraints, fetch data, and resolve codes. No obvious gaps for a public statistics API.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides seamless access to official Ukrainian government statistics from the State Statistics Service via their SDMX API. It enables users to query bilingual datasets across domains like energy, demographics, and trade using flexible filtering and metadata exploration tools.
      5 npm
      50
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables querying UK Office for National Statistics datasets and their editions through natural language, with no authentication required.
      72 npm
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      Provides access to UK economic, social, and labour-market statistics via the ONS beta and Nomis APIs, enabling discovery and retrieval of time series, datasets, and census data.
      18
      MIT

    Appeared in Searches