Tdcv2

TDCv2 for AI agents

A data generator for your coding agent: fake, mock, synthetic and test data with TDCv2 — realistic names, emails, addresses, phone numbers, IBANs and more. The agent writes a small .tdc config, the TDCv2 engine generates the rows on your machine, and the same config gives the same file, byte for byte, every time you run it. CSV, JSON, SQL inserts or any text format; fields that agree with each other; exact shares; related tables; data packs for 95 languages and 198 countries. Think Faker, but for whole datasets: database seeds, test fixtures, demo data.

Two parts, one version:

• a skill — teaches the agent the TDC language, the workflow and the traps (skills/tdcv2);
• an MCP server, tdcv2-mcp — the same, as tools: check a config, generate, find and install data packs, summarise a file, read the docs (mcp/).

Install

Claude Code and Cowork — the plugin brings both the skill and the server. The simplest way: in claude.ai or the Claude app, Customize → Plugins → Discover, find TDCv2 — Fake & Test Data Generator, Add; Claude Code picks it up at the next session and keeps it updated. Or from this repository's marketplace:

claude plugin marketplace add NickLiapin/tdcv2-agents

claude plugin install tdcv2@tdcv2

Install one way, not both — otherwise the skill loads twice and two servers start.

Codex, Copilot, Gemini CLI, Cursor and other agents that read skills — put the skill where your agent looks for it:

npx -y tdcv2-mcp@0.1.7 install-skill            # ~/.claude/skills/tdcv2

npx -y tdcv2-mcp@0.1.7 install-skill --agents # ~/.agents/skills/tdcv2 npx -y tdcv2-mcp@0.1.7 install-skill --project # ./.claude/skills/tdcv2 npx -y tdcv2-mcp@0.1.7 install-skill --dir

Any MCP client (Claude Desktop, Cursor, VS Code…):

{ "mcpServers": { "tdcv2": { "command": "npx", "args": ["-y", "tdcv2-mcp@0.1.7"] } } }

Claude Desktop starts servers in /: add "env": {"TDCV2_PROJECT": "/path/to/project"} to that entry so the files have a folder to go to. On Windows use "command": "npx.cmd" if the client cannot start npx. Everything needs Node.js 20 or newer.

Step by step for each agent and client: TDC for AI agents.

Then ask in plain words — *"500 customers from 20 countries as CSV"*, *"a pytest test for validate_user on 50 different users"* — and the agent takes it from there: writes the config, checks it with the engine, generates, and hands you the file together with the config that reproduces it.

Privacy

The engine runs locally. 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 project folder it is started in.

Network access, all of it:

• the plugin starts the server with npx -y tdcv2-mcp@0.1.7, its own version (scripts/mcp-launch.mjs), so the first start downloads that exact version and its dependencies, the TDCv2 engine among them, from the npm registry;
• npx -y tdcv2-mcp@0.1.7 … in the commands above does the same, and so does npx -y tdcv2@0.3.7 …, the exact engine version the agent runs;
• with an engine of another version, the agent may read that version's doc page from raw.githubusercontent.com;
• data packs are downloaded only when you or the agent run tdcv2 pack add (through the server: tdc_pack_add, which your client asks you to allow);
• a config with sends the named values to that URL — only if someone writes such a config.

Nothing else is fetched, and nothing is sent. The full policy: PRIVACY.md.

Versions

This release, 0.1.7, is checked against TDCv2 0.3.7. The skill's reference is built from the engine's documentation for that version; with another engine version the agent reads that version's pages, and the engine's own check has the last word.

License

MIT