Build on the world's most comprehensive classification graph

Popular endpoints

The MCP (Model Context Protocol) server lets AI assistants like Claude query the taxonomy graph directly - searching codes, translating between systems, navigating hierarchies, and exploring country profiles - all from within a conversation.

Quick start

# From the repo root (requires DATABASE_URL in environment)
python3 -m world_of_taxonomy mcp

Popular tools

Four packaged integrations, all backed by the same REST API and MCP server. Pick the one that matches your agent runtime. Source lives in the /skills directory of the repo.

Shared capabilities

• Classify free text (business, product, occupation, document) under standard codes across all systems
• Translate codes between systems (NAICS -> ISIC, ICD-10-CM -> ICD-10-GM, SOC -> ISCO, HS -> CPC)
• Walk hierarchies (children, ancestors, siblings) and audit crosswalk coverage between any two systems

Every system follows the same TDD cycle: write a failing test first, implement the ingester to make it green, wire it into the CLI, then run the full suite to confirm no regressions. The detailed SOP lives in docs/adding-a-new-system.md.

10-step checklist

1. 1Write a failing test (test_ingest_<system>.py) - confirm it is red before continuing
2. 2Create the ingester (ingest/<system>.py) - parse source data, build SYSTEM + NODES dicts
3. 3Set is_leaf correctly - use codes_with_children = {parent for ... if parent} pattern
4. 4Implement ingest(conn) - upsert system row, upsert nodes in dependency order
5. 5Run the test green - minimum code to pass, nothing more
6. 6Add crosswalk edges if a concordance table exists (ingest/crosswalk_<system>.py)
7. 7Write a test for equivalences - confirm bidirectional edges are created
8. 8Wire into __main__.py ingest command (add system id to the dispatch table)
9. 9Run the full test suite - python3 -m pytest tests/ -v - all green before committing
10. 10Update CLAUDE.md system table with name, region, and node count

Three implementation paths

Minimal standalone ingester template

# world_of_taxonomy/ingest/my_system.py
SYSTEM = {
    "id": "my_system_2024",
    "name": "My Classification System 2024",
    "authority": "Issuing Body",
    "region": "Global",
    "version": "2024",
    "description": "...",
}

# (code, title, description, parent_code)
NODES = [
    ("A", "Section A", "Agriculture", None),
    ("A01", "Crop production", "...", "A"),
    ...
]

async def ingest(conn) -> None:
    await conn.execute("""
        INSERT INTO classification_system (...) VALUES (...)
        ON CONFLICT (id) DO UPDATE SET ...
    """, *SYSTEM.values())

    # Compute leaf flags dynamically - never hard-code level == N
    codes_with_children = {parent for (_, _, _, parent) in NODES if parent is not None}

    for code, title, desc, parent in NODES:
        is_leaf = code not in codes_with_children
        await conn.execute("""
            INSERT INTO classification_node (...) VALUES (...)
            ON CONFLICT (system_id, code) DO UPDATE SET ...
        """, SYSTEM["id"], code, title, desc, parent, is_leaf)