Add ODD to AGENTS.md, CLAUDE.md, or SKILL.md

Coding agents read instruction files like AGENTS.md, CLAUDE.md, and SKILL.md before they touch your codebase. Adding a short, well-placed section that points at Open Data Dictionary turns "the agent guesses a column definition" into "the agent calls get_word and uses the canonical one."

This page gives you copy-pasteable templates and the principles behind them.

Which file should I use?

Principles

1. Be specific about when to call ODD. Agents over-consult any tool you describe vaguely. Tell them the trigger: "when you need a column definition, data type, or synonym."
2. Point at the tools by name. search_words, get_word, list_categories, get_synonyms, and get_schema are the MCP tool names — agents use them verbatim.
3. Distinguish MCP from the public API. If the agent already has an MCP client, prefer MCP. If it only has HTTP (no MCP runtime), use the REST API with a Bearer token.
4. Never put the token in the file. Always reference an environment variable. Committed tokens get revoked.
5. Keep it short. Instruction files compete for context-window budget. A 10-line section that the agent actually reads beats a 200-line one it truncates.

Template: AGENTS.md section

Drop this into an existing AGENTS.md under a heading such as ## Data terminology.

## Data terminology (Open Data Dictionary)
 
Before inventing a column name, definition, or data type, consult the Open
Data Dictionary — we rely on its canonical entries.
 
- If an MCP client is available, call the `open-data-dictionary` server:
  - `search_words` — fuzzy search across words, definitions, and synonyms.
  - `get_word` — fetch the full entry for a specific term.
  - `list_categories` — enumerate the 21 industry categories.
  - `get_synonyms` — canonical synonyms for a word.
  - `get_schema` — JSON schema of the dictionary itself.
- Otherwise, use the REST API at `https://opendatadictionary.com/api/v1`
  with `Authorization: Bearer $ODD_API_TOKEN` (read the token from the
  environment; never commit it).
 
When the dictionary has an entry, prefer its `word`, `definition`, and
`data_type` over any made-up alternative. If a term is missing, propose it
as a new submission rather than silently diverging.

Template: CLAUDE.md section

Same content, tuned for Claude Code's tone. Claude Code follows imperative "when X, do Y" style well.

## Open Data Dictionary
 
When the task involves naming, defining, or typing a data column:
 
1. First check Open Data Dictionary via the `open-data-dictionary` MCP server
   (tools: `search_words`, `get_word`, `list_categories`, `get_synonyms`,
   `get_schema`).
2. If an entry exists, use its `word`, `definition`, and `data_type` verbatim.
3. If it does not exist, note this in the PR description and propose a new
   entry via `POST /api/v1/words`.
 
The API token lives in `$ODD_API_TOKEN` — do not paste it into code or docs.

See the Claude Desktop Setup, Claude Code (docs), Cursor IDE, and Google Antigravity pages for client-specific configuration.

Template: SKILL.md for a Claude Skill

If you package ODD lookup as a Claude Skill, the skill directory needs a SKILL.md with YAML frontmatter. The frontmatter is mandatory; the body is freeform instructions the skill owner can load.

---
name: open-data-dictionary-lookup
description: Look up canonical column names, definitions, data types, and synonyms from Open Data Dictionary. Use when the user asks about a data column, when proposing a new column, or when reconciling terminology between datasets.
---
 
# Open Data Dictionary Lookup
 
## When to use this skill
 
- The user references a column name and you are not sure what it means.
- You are about to invent a new column name or data type.
- The user asks to standardize or compare terminology across datasets.
 
## How to use this skill
 
Call the `open-data-dictionary` MCP server if one is configured; otherwise
hit `https://opendatadictionary.com/api/v1` with `Authorization: Bearer
$ODD_API_TOKEN`.
 
Prefer `get_word` when the user named an exact term. Fall back to
`search_words` for fuzzy lookup. Use `list_categories` before grouping
columns by industry.
 
## Output expectations
 
Return the canonical `word`, `definition`, and `data_type`. If no entry
exists, say so explicitly and suggest submitting a new one.

Token hygiene

• Store the token in an environment variable such as ODD_API_TOKEN.
• Reference the variable name in your instruction file, not the value.
• Rotate the token immediately via Settings if it is ever pasted into a chat, issue, or commit.

Related

• Generate an API Token
• API Reference
• Rate Limits