mcs v0.18
// MaxCompute semantic skill for AI agents / CLI + SKILL.md

Take MaxCompute
your AI agent

Semantic package + memory store — the more your AI Agent uses it, the better it understands your MaxCompute.

// install
agent> Install mcs for me, read this guide fully then follow step by step: curl -fsSL https://raw.githubusercontent.com/aliyun/maxcompute-semantic/main/scripts/install.md
[00]Concept /03

Three nouns. Every command revolves around them.

You configure a profile mcs produces a semantic package agent reads it to write SQL

[A]profile
  • authAK / process
  • scopesources × N
[B]semantic package
  • analyzesample + metadata
  • inferJOIN / UDF
[C]agent
  • readlocal package
  • runcost → execute

scenario = [A] + [B] + accumulated annotate / memory

Read more → docs

[01]Demo /02

The agent invokes mcs via a skill using natural language; you can also open a terminal and run mcs commands directly.

user/maxcompute-semantic What was the GMV from the orders table last month? skill loaded · maxcompute-semantic mcssql execute "SELECT ds, SUM(amount) ..."
[02]Core features /06

Give your AI Agent a colleague who knows MaxCompute.

[01]

Cross-agent skill bundle

A single mcs skill install --all symlinks SKILL.md into Claude Code / Cursor / Codex / Gemini / Qwen / OpenCode and 50+ more platforms. All agents auto-sync when mcs upgrades.

[02]

Automatic semantic profile build

mcs build scans and indexes tables / columns / JOINs / UDFs / annotations in one pass. The agent reads the result before generating SQL.

[03]

Dialect-aware SQL + cost gate

Generates SQL in MaxCompute dialect. mcs sql cost gates execution against a configurable threshold before running.

[04]

Query memory + hybrid retrieval

mcs memory accumulates query experience. Similar questions are recalled via inverted-index + vector hybrid retrieval — accuracy improves with use.

[05]

One-click install / auto-update

Install with one command. New versions are auto-detected and prompted. mcs update upgrades in place.

[06]

mcs doctor self-check

Covers config / auth / network / skill checks. -f json output lets agents parse results automatically.

[03]Next steps /03

What to do after installation. Every command has full --help examples.

01

From zero to first query

Tell your agent "install mcs for me" — it follows install.md to set up the CLI and skill. Then ask any MaxCompute data question; the skill guides you through auth and query execution.

02

What shipped recently

See what each version adds, changes, and fixes. The changelog is synced automatically by the release pipeline.

release notes →
03

Feedback / report bugs

In your AI agent, say "report an mcs bug" — the skill walks through the full issue-filing workflow.