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., "@International Classifications MCP find the ISCO-08 code for software developers"
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.
A free, open-source, deterministic MCP server for discovering, explaining, validating and mapping official international statistical classifications.
It is designed for questionnaire development, survey coding, statistical harmonisation and reproducible analysis. No LLM, embedding service or paid AI runs inside the server. The connected AI client interprets the user's task; this service provides versioned official structures, deterministic retrieval and provenance.
Production safeguards include first-class typed codelist hierarchy nodes, deterministic routing fixtures for every curated codelist, weekly authoritative-link checks, per-tool request IDs and latency logging, and documented OpenAI directory annotation justifications.
The registry includes full machine-readable structures for ISIC Rev.5, CPC 3.0, COICOP 2018, HS 2022, SITC Rev.4, BEC Rev.5 and UN M49; a complete ISCO-08 code/title hierarchy; the official ISIC Rev.4 to Rev.5 correspondence; operational ICSE-18, ICSaW-18 and ISCED structures; detailed ISCED-F and ICCS structures; and clearly labelled reference-only entries for ICD-11 and ICF.
MICS7 is represented only by curated, question-specific response codelists for matching questionnaire answer categories. MICS modules and analytical indicator definitions are intentionally not exposed as classifications; indicator discovery belongs in a development-indicators service. The response layer does not replace UNICEF's current questionnaires or country customisation guidance and never silently combines MICS rounds.
Version 0.5 adds ICC 1.1, ICATUS 2016, COFOG, COPNI, COPP, UNECE Recommendation 20 units and Recommendation 21 package types; detailed ISCED-F and ICCS titles; expanded ICSE/ICSaW and SDMX cores; and separately versioned DHS-8, Washington Group, JMP WASH, WHO vaccination and FAO WCA questionnaire codelists.
Each curated answer list has a stable codelist_id. Call list_codelists, then pass one identifier to search or export. The server refuses to merge unrelated MICS choices into one questionnaire list.
National census and administrative geography codes are intentionally out of scope because they change frequently and lack one authoritative global registry.
Coverage is explicit in list_classifications: full, seed, mapping, reference, or curated. Reference-only and curated families are never presented as complete code lists.
Related MCP server: istat-mcp-suite
list_classifications
get_classification
list_codelists
recommend_classifications
search_codes
get_code_definition
browse_hierarchy
validate_codes
map_codes
export_choice_list
codex mcp add international-classifications --url https://classifications.impactengines.ai/mcpThe root URL provides service status; /mcp is the MCP endpoint.
python -m pip install -e ".[dev]"
python -m international_classifications_mcp.build_registry
pytest
classifications-mcpFor HTTP transport:
MCP_TRANSPORT=streamable-http MCP_HOST=0.0.0.0 classifications-mcpOfficial custodians remain authoritative.
Source files are hashed and registry releases are versioned.
Search is programmatic: FTS5, curated concept rules and explicit scoring.
Correspondences preserve one-to-many and changed-definition warnings.
Detailed occupations, industries, diseases and crimes are normally post-coded, not dumped into respondent-facing questionnaires.
Source content remains subject to its custodian's terms; the MIT licence covers this software, not third-party classification content.
See source audit and agent test prompts.
Browse top-level nodes or immediate children. Curated families return node_kind=codelist containers at root; pass that code as parent_code to browse its options.
Inputs: classification_id (required), limit, parent_code. Returns: result
Return a structured choice list at a selected hierarchy level. MICS exports require one codelist_id and never merge unrelated answer lists.
Inputs: classification_id (required), level, limit, format, +1 more. Returns: rows, level, format, warning, +4 more
Get authoritative metadata and coverage for one classification_id returned by list_classifications.
Inputs: classification_id (required). Returns: id, name, domain, status, +8 more
Retrieve one exact code or codelist container, including its node_kind, hierarchy position and explanatory notes.
Inputs: code (required), classification_id (required). Returns: code, label, level, score, +11 more
List classification families, versions, custodians, coverage and licensing notes. Start here when the correct standard is unknown.
Inputs: domain, status. Returns: result
List independent option lists within a classification before searching or exporting one list.
Inputs: classification_id (required). Returns: result
Apply official stored correspondences and expose splits, merges, changed meanings and unmapped codes. Never assume one-to-one equivalence.
Inputs: codes (required), source_classification_id (required), target_classification_id (required). Returns: results, sources, warnings, source_classification, +1 more
Return a deterministic ranked shortlist of all relevant classification concepts and codelists. Ranking supports user choice; it is not an automatic selection. Include answer options and context.
Inputs: question_text (required), limit, answer_options, survey_context. Returns: warnings, next_action, recommendations
Full-text search labels and notes. Results restrict to one option list only when codelist_id is explicitly supplied.
Inputs: query (required), limit, codelist_id, classification_ids. Returns: query, total, results, warnings
Validate exact codes against one named classification version.
Inputs: codes (required), classification_id (required). Returns: results, valid_count, invalid_count, classification_id
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 10 tools
Each tool targets a distinct operation in the classification workflow: listing families, listing option lists, metadata retrieval, recommendation, search, code lookup, hierarchy browsing, validation, mapping, and export. The only mild overlap is between recommend_classifications and search_codes, but their descriptions clearly separate ranked concept selection from full-text label/note search.
All tool names use snake_case with a consistent verb_noun structure, such as list_classifications, get_classification, search_codes, validate_codes, and map_codes. Minor variations in token length do not break the predictable pattern.
The server exposes 10 tools, which is well within the ideal range for a focused reference-data MCP. Each tool earns its place by covering a distinct discovery, retrieval, validation, mapping, or export need without redundant CRUD operations.
The surface covers the full read-only lifecycle for international classifications: discovery, metadata, recommendation, search, hierarchy browsing, exact code retrieval, validation, mapping, and choice-list export. No create/update/delete tools are expected for authoritative reference standards, and any bulk navigation is handled by browse_hierarchy and export_choice_list.
UK Office for National Statistics dataset catalogue + Beta JSON API
Classification service to seamlessly retrieve SOC codes for job titles and descriptions.
SDMX Structure MCP — dataflow dimensions and code lists for any SDMX 2.1 registry: Eurostat, OECD…
Search, filter, count and sum Australian government open data, with every version kept.
