World Of Taxonomy

MCP Setup Guide for Claude Desktop, Cursor, Zed, and other AI clients

MCP setup guide

TL;DR: Get an API key at worldoftaxonomy.com/developers, paste it into your AI client's MCP config, and ask the assistant taxonomy questions in plain English. Five minutes start to finish.

The World Of Taxonomy MCP server gives AI assistants (Claude Desktop, Cursor, Zed, VS Code Continue, and any other MCP client) direct access to 1,000+ classification systems, 1.3M+ codes, and 321K+ crosswalk edges. Once installed, you can ask things like "convert NAICS 5415 to NACE" or "find ICD-10 codes related to type 2 diabetes" and get authoritative answers without leaving the editor.

The server is published as the worldoftaxonomy-mcp package on PyPI. End users do not need a database, a clone of the repo, or any local build step - the package handles every tool call by hitting the WoT REST API with your key.

Step 1: Get your API key

  1. Visit worldoftaxonomy.com/developers.
  2. Enter your email and click Send me a sign-in link.
  3. Open the email and click the link. The dashboard loads.
  4. Click Generate new key. Name it after where you'll use it (e.g., MCP on laptop).
  5. Copy the key shown on screen. It's shown only once.

The key looks like wot_a3f2c5d9b8e7f6c4d2a1b0c9d8e7f6a5 (free-tier keys start with wot_; restricted or cross-product keys may use rwot_ or aix_).

Free tier gives you 1,000 requests per minute shared across your team. No credit card. See API key management for rotation, revocation, and scoping.

Step 2: Install for your AI client

Pick the client you use. Each has its own config file location.

Claude Desktop

Mac: ~/Library/Application Support/Claude/claude_desktop_config.json

Windows: %APPDATA%\Claude\claude_desktop_config.json

Add this to the mcpServers block (create the block if it doesn't exist):

{
  "mcpServers": {
    "worldoftaxonomy": {
      "command": "uvx",
      "args": ["worldoftaxonomy-mcp"],
      "env": {
        "WOT_API_KEY": "wot_a3f2c5d9b8e7f6c4d2a1b0c9d8e7f6a5"
      }
    }
  }
}

Restart Claude Desktop (Cmd-Q, relaunch). You'll see a small hammer icon at the bottom of the chat input - that's MCP. Click it to confirm worldoftaxonomy appears in the tool list.

If you don't have uvx installed and don't want to install it, the plain-pip alternative works too:

pip install --user worldoftaxonomy-mcp

Then change command to worldoftaxonomy-mcp and drop args. The uvx path is recommended because it isolates the install in a per-invocation venv and uses the latest published version automatically.

Cursor

Cursor uses MCP via Settings -> Features -> Model Context Protocol.

  1. Open Settings -> Features -> Model Context Protocol.
  2. Click Add new MCP server.
  3. Enter:
    • Name: worldoftaxonomy
    • Type: stdio
    • Command: uvx worldoftaxonomy-mcp
  4. Add an environment variable: WOT_API_KEY = your key.
  5. Save and restart Cursor.

Zed

Zed uses ~/.config/zed/settings.json. Add to context_servers:

{
  "context_servers": {
    "worldoftaxonomy": {
      "command": {
        "path": "uvx",
        "args": ["worldoftaxonomy-mcp"],
        "env": {
          "WOT_API_KEY": "wot_a3f2c5d9b8e7f6c4d2a1b0c9d8e7f6a5"
        }
      }
    }
  }
}

Restart Zed.

VS Code (with Continue extension)

Continue's MCP support lives in ~/.continue/config.json:

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "uvx",
          "args": ["worldoftaxonomy-mcp"],
          "env": {
            "WOT_API_KEY": "wot_a3f2c5d9b8e7f6c4d2a1b0c9d8e7f6a5"
          }
        }
      }
    ]
  }
}

Restart the Continue extension (Cmd-Shift-P -> "Continue: Restart").

Windsurf

Windsurf uses ~/.codeium/windsurf/mcp_config.json with the same schema as Claude Desktop:

{
  "mcpServers": {
    "worldoftaxonomy": {
      "command": "uvx",
      "args": ["worldoftaxonomy-mcp"],
      "env": {
        "WOT_API_KEY": "wot_a3f2c5d9b8e7f6c4d2a1b0c9d8e7f6a5"
      }
    }
  }
}

Restart Windsurf.

Generic MCP client

If your client is a generic MCP host:

  • Transport: stdio
  • Command: uvx worldoftaxonomy-mcp (or worldoftaxonomy-mcp if pip-installed)
  • Required env var: WOT_API_KEY
  • Optional env var: WOT_API_BASE_URL (defaults to https://wot.aixcelerator.ai)
  • Protocol version: MCP 2024-11-05 or later

Step 3: Verify it works

Open a chat with your AI client and try one of these prompts:

"Look up NAICS 2022 code 5417 and tell me what NACE Rev 2 code it maps to."

"Find ICD-10-CM codes related to type 2 diabetes mellitus."

"Translate ISIC Rev 4 division 84 to its equivalent in the German WZ 2008 classification."

"What classification systems are commonly used in Brazil?"

The assistant should call MCP tools (you'll see something like get_industry, translate_code, or get_equivalences in the tool log) and answer with specific codes and titles, not generic AI reasoning.

Available tools

The MCP server exposes 26 tools. The most commonly used:

Tool Use it for
list_classification_systems "What systems are available?"
search_classifications Full-text search across all codes
get_industry Look up a specific code
browse_children Drill into a category
get_equivalences Get all crosswalk mappings for a code
translate_code Convert between two specific systems
translate_across_all_systems Convert to every connected system
classify_business Free text -> taxonomy codes
get_country_taxonomy_profile Systems applicable in a country

The full list with arguments and return types is in the API reference at worldoftaxonomy.com/api.

A handful of tools (list_crosswalks_by_kind, get_country_scope, get_audit_report) are available in self-hosted mode but not yet wired into the published PyPI package; they'll surface a "not yet supported" error if invoked. They are not on the critical path for typical AI-agent tool calls. Open an issue if you have a specific need.

Troubleshooting

"WOT_API_KEY not set" on startup

The MCP server requires the key as an environment variable. Re-check the env block in your client's config. Mac/Linux users can verify the wiring with a one-shot stdin probe:

echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
  | WOT_API_KEY=wot_a3f2c5d9... uvx worldoftaxonomy-mcp \
  | head -c 200

You should see a JSON-RPC response that starts with {"jsonrpc": "2.0", "id": 1, "result": {"protocolVersion":...}. If you get an exit-2 stderr message about "no credentials", the env var is not actually being read; double-check the env block in your client's config (not all clients allow env-var inheritance from the parent shell).

"Authentication failed" / 401 errors

Your key may be revoked or expired. Sign in at worldoftaxonomy.com/developers, check the key's status in the dashboard. Generate a new one if needed; both old and new work simultaneously, so update your mcp.json first, then revoke the old.

"Rate limit exceeded" / 429 errors

You or your team has hit the free-tier 1,000 req/min ceiling. Options:

  • Wait 60 seconds and retry. The bucket refills.
  • If you're on a corporate domain with multiple developers sharing the pool, ask the org admin who's burning the budget.
  • Upgrade to the Pro tier (10,000 req/min pool) at worldoftaxonomy.com/pricing.

Tools don't appear in the AI client

  1. Restart the client fully (quit and relaunch, not just close the window).
  2. Verify the config file is valid JSON. A trailing comma or missing brace silently disables MCP.
  3. Check the client's MCP log:
    • Claude Desktop: View -> Developer -> View Logs from MCP servers.
    • Cursor: Cmd-Shift-P -> "Cursor: Show Logs" -> filter for mcp.
    • Zed: Cmd-Shift-P -> "zed: open log" -> search for worldoftaxonomy.

"Command not found: uvx"

Install uv (the runner that ships uvx):

  • Mac / Linux: curl -LsSf https://astral.sh/uv/install.sh | sh
  • Windows: irm https://astral.sh/uv/install.ps1 | iex

Or skip the runner entirely and pip install --user worldoftaxonomy-mcp, then point your client at worldoftaxonomy-mcp directly (no args needed). The uvx path is preferred because it runs the latest published version on every invocation; pip install pins you to whatever version was current at install time.

Server crashes when answering certain questions

Capture the error from the client's MCP log and report it via the Report incorrect description link on the relevant node detail page, or open an issue on the GitHub repo. Include the exact prompt and the MCP log excerpt.

What's next