TDC for AI agents

A coding agent can write TDC configs for you. You ask in plain words — "500 customers from 20 countries as CSV", "a pytest test for validate_user on 50 different users" — and the agent writes the .tdc config, checks it with the engine, generates the file on your machine, and hands you the data together with the config that reproduces it.

An agent needs two things for that, and the tdcv2-agents project supplies both:

• A skill — instructions the agent reads before it starts. It teaches the TDC language, the workflow, and the traps the engine's check cannot catch. The agent then runs the engine in its own terminal.
• An MCP server, tdcv2-mcp — the same knowledge as tools the agent calls: check a config, generate, find and install data packs, summarize a file, read these docs. MCP (the Model Context Protocol) is the standard way an agent calls tools that another program provides.

Both ship together, under one version: 0.1.6.

Which one to use

Your agent	Install	Page
Claude Code	the plugin, which brings both	Claude Code
Codex, Cursor, VS Code with Copilot, Gemini CLI	the skill; add the server if you want both	Other agents
any other MCP client	the server	Other agents

The skill is enough for an agent that has a terminal. The server suits a client without one, or an agent that works better with tools than with instructions. Installing both does no harm: they teach the same thing.

Where it is published

Where	Name
Anthropic's plugin directory	TDCv2 — Fake & Test Data Generator, in claude.ai and the Claude app under Customize → Plugins → Discover
Source code (MIT)	NickLiapin/tdcv2-agents
Claude Code marketplace	NickLiapin/tdcv2-agents, plugin tdcv2@tdcv2
Official MCP Registry	io.github.NickLiapin/tdcv2
npm	tdcv2-mcp — the server and the skill

Everything needs Node.js 20 or newer.

Versions

Release 0.1.6 is checked against TDC 0.3.6. The skill's reference is built from that engine's documentation, and the agent runs that engine: npx -y tdcv2@0.3.6. The server prefers your project's own tdcv2 when the project has one, and otherwise uses the engine it ships with. With another engine version, the agent reads that version's pages, and the engine's check has the last word.

Privacy

The engine runs on your machine. Neither the skill nor the server sends your data or your configs anywhere. The server reads .tdc files and writes generated files only inside the folder it is started in.

The network is used for four things only:

• the first start downloads tdcv2-mcp and the engine from npm;
• with an engine version other than 0.3.6, the agent may read that version's documentation page from GitHub;
• data packs download when you or the agent run tdcv2 pack add — through the server, that is tdc_pack_add, which your client asks you to allow;
• a config with <gen type="http" src="…"> sends the named values to that address, and only if someone writes such a config.

The full policy is in PRIVACY.md.