Your Cursor already knows Treasure AI

Query customer data, build segments, and run workflows with Cursor’s Agent — right in the editor you already live in, with production-grade governance. Two minutes to connect.

90%

Less time on routine CDP work.

2 minutes

from setup to your first AI-built segment.

$0

works with the Cursor plan your team already has.

Quick start

Two minutes in your existing Cursor

Install the tdx CLI, add its MCP server to mcp.json, and ask the Agent. Your Cursor — your plan, your settings — now operates your CDP.

Set up your editor

Connect Treasure AI to Cursor

The tdx MCP server gives Cursor’s Agent two clean tools — tdx_run and tdx_search — covering the entire platform.

# 1 · install & authenticate the CLI npm install -g @treasuredata/tdx tdx auth setup # 2 · add the MCP server — ~/.cursor/mcp.json { "mcpServers": { "tdx": { "command": "tdx", "args": ["mcp"] } } } # 3 · open Agent chat and ask > “Which of our databases has last week’s web events?”
  • Works with the Cursor plan you already have
  • Also configurable from Cursor Settings → MCP
  • Respects your TD roles and policies on every call
Roll it out to the team

One file, whole team connected

Commit .cursor/mcp.json to your repo and everyone who opens the project gets Treasure AI in the Agent — no per-machine setup.

// .cursor/mcp.json — commit to your repo { "mcpServers": { "tdx": { "command": "tdx", "args": ["mcp"] } } }
  • Each developer authenticates as themselves with tdx auth setup
  • No global install needed — use "command": "npx", "args": ["@treasuredata/tdx", "mcp"]
  • Project-scoped, so different repos can point at different TD profiles
Level up

Add TD Skills to Cursor

TD Skills are the CDP playbooks written by the people who built the platform — segment YAML patterns, Trino optimization, workflow debugging. Cursor reads the open Agent Skills format natively: drop them in and run with /skill-name, or let the Agent pick the right one.

# clone the library, copy the skills you want git clone https://github.com/treasure-data/td-skills mkdir -p ~/.cursor/skills cp -r td-skills/tdx-skills/* td-skills/sql-skills/* ~/.cursor/skills/ # team rollout: commit them to .cursor/skills/ in your repo instead
  • Run a skill with /skill-name, attach with @ — or let the Agent auto-load the match
  • Open Agent Skills format — every skill is a readable Markdown playbook, not a black box
  • The same skills power Claude Code, Copilot, and Codex — one library, every editor
TD Skills

Playbooks, not prompts

TD Skills are CDP playbooks in the open Agent Skills format, written by the people who built the platform. Copy what you need into ~/.cursor/skills/ and run with /skill-name — the same library that powers Claude Code, Copilot, Codex, and Treasure Code.

tdx-skillsSegments, parent segments, journeys, activations, Engage campaigns, Foundry agents
sql-skillsTD-flavored Trino & Hive, query optimization, time filtering
workflow-skillsDigdag pipelines, dbt, LLM-powered workflows
realtime-skillsReal-time personalization, triggers, ID stitching
sdk-skillsJavaScript & Python SDK integration patterns
creative-skillsAd ideation, brand compliance, email & social creatives
studio-skillsTreasure AI Studio workspace and scheduling
field-agent-skillsDeployment, documentation, visualization

Browse all skills on GitHub →  ·  MCP server docs →

What you can do

Real CDP work, in plain language

Not a chatbot bolted onto a dashboard — Cursor’s Agent drives the same tdx commands your team uses, so everything it makes is reviewable and reproducible.

⌕

Explore customer data

Browse databases, inspect schemas, and run Trino/Hive queries without leaving your editor.

Which tables hold purchase events, and how fresh are they?
◎

Build & ship segments

Draft segment rules as YAML, validate match rates, and push to production with an audit trail.

Build a segment of lapsed high-LTV buyers in Japan
⚙

Debug workflows

Diagnose failed Digdag runs, patch the .dig file, and retry — with the error context pulled in automatically.

Why did last night’s enrichment workflow fail?
↦

Design journeys

Compose multi-stage journeys with decision points and A/B tests, validated before anything goes live.

Add a 3-day wait and a coupon branch to the winback journey
✉

Create campaigns

Generate Treasure Engage email templates that pass brand and compliance checks on the first try.

Draft the July renewal email for the cart-abandoner segment
❏

Find the right command

The Agent uses tdx_search to discover the exact tdx command for the task — no manual paging through references.

What’s the command to preview a parent segment?
How it works

No side doors — just tdx, over MCP

The Agent never touches raw credentials or a private API. Every action flows through the tdx MCP server as a real tdx command — one you could type yourself, authenticated as you, governed and logged like any other.

Cursor Agent The Agent plans the task and calls only the tools you’ve granted it.
→
tdx MCP server Translates requests into plain tdx commands — validated before anything is pushed.
→
Treasure AI platform CDP, segments, journeys, workflows — with your org’s policies enforced.

Prefer Claude Code, VS Code, or Codex? See Works with Claude Code, Works with VS Code, and Works with Codex.

Policy-scopedAgents inherit the caller’s TD roles — nothing more.
Fully auditedEvery command lands in the audit log, same as the console.
Version-controlledSegments and workflows are YAML — diffable, reviewable, revertible.
Private by defaultYour customer data is never used to train models.

FAQs

Treasure AI doesn’t require any particular Cursor plan — the tdx CLI and its MCP server are free, and platform usage is covered by your Treasure AI contract. Agent usage itself is billed by Cursor under the plan your team already has.

Actions run through tdx with your TD role and policies — the agent can’t do anything you couldn’t do yourself. Auth and profile commands are blocked over MCP by design, everything is audit-logged, and segment or workflow changes are plain YAML you can review before pushing.

Databases and SQL (Trino/Hive), parent segments and segments, journeys, activations, Digdag workflows, Treasure Engage campaigns, and Foundry agents — everything the tdx CLI covers, exposed to the Agent through the tdx_run tool.

Yes — that’s the point. Marketers describe the segment or journey they want in plain language; the YAML, validation, and push happen underneath. For a fully no-code experience outside the editor, see Treasure AI Studio.

Yes — Cursor supports the open Agent Skills format natively. Copy skills into ~/.cursor/skills/ (personal) or .cursor/skills/ (project) from treasure-data/td-skills, then run them with /skill-name or attach with @ — or let the Agent pick the right one automatically. Every skill is a readable Markdown playbook you can audit before installing.

The tdx CLI and its MCP server are free. Platform usage is covered by your existing Treasure AI contract, and Cursor is billed under the plan your team already has — nothing new to procure.