Claude Code & Codex

Use Claude Code or OpenAI Codex to work with your rawquery data through natural language. Ask questions about your data, write SQL, build dashboards. The agent handles everything for you.

The setup is the same in both: install the rq CLI, log in, drop a skill in your project. Same workspace, same data — just two different agents driving it.

Setup

Option A: CLI (recommended)

The rq CLI is the core rawquery interface. Install it, log in, then generate the skill for your agent.

For Claude Code:

curl -sSL rawquery.dev/install.sh | sh
rq login
rq claude-init

curl -sSL rawquery.dev/install.sh | sh
rq login
rq claude-init

Creates .claude/skills/rq/SKILL.md in your current directory. Claude Code picks it up automatically.

For Codex:

curl -sSL rawquery.dev/install.sh | sh
rq login
rq codex-init

curl -sSL rawquery.dev/install.sh | sh
rq login
rq codex-init

Creates .agents/skills/rq/SKILL.md plus an agents/openai.yaml config so Codex loads the workspace context on start.

In both cases the agent runs rq commands on your behalf with the same auth, workspace, and data.

Option B: MCP server (no CLI needed, Claude Code only)

For environments where you can't install the CLI (Cowork, remote setups), connect to the hosted MCP server:

{
  "mcpServers": {
    "rawquery": {
      "type": "sse",
      "url": "https://mcp.rawquery.dev/sse",
      "headers": {
        "X-API-Key": "rq_your_key_here",
        "X-Workspace": "your-workspace-slug"
      }
    }
  }
}

{
  "mcpServers": {
    "rawquery": {
      "type": "sse",
      "url": "https://mcp.rawquery.dev/sse",
      "headers": {
        "X-API-Key": "rq_your_key_here",
        "X-Workspace": "your-workspace-slug"
      }
    }
  }
}

Get your API key at rawquery.dev > Settings > API Keys. No CLI, no local dependencies. Just your API key.

You can also install the rawquery plugin for a skill, agent, and guided commands on top of the MCP tools:

/plugin install github:rawquery/rawquery-plugin

What you get

With the CLI (Option A)

The agent runs rq commands on your behalf. The skill loads your workspace context dynamically: schemas, tables, connections, saved queries, transforms, charts, and pages. The agent always sees your current state. Works the same in Claude Code and Codex.

With the MCP server (Option B, Claude Code only)

Claude gets 20 tools for direct API access: list_tables, describe_table, execute_query, create_saved_query, create_chart, create_page, publish_page, and more.

The plugin adds on top of that:

Examples of what you can ask:

"Show me my top 10 customers by revenue"
"Join my Stripe charges with HubSpot contacts"
"What tables do I have?"
"Create a chart showing monthly revenue trends"
"Sync my Stripe connection"

Workflows

Ask a question about your data

Just ask in natural language. The agent explores your schema, writes SQL, runs the query, and presents results.

You: "What's my MRR by month for the last 6 months?"

Agent: describes stripe.charges
       executes SELECT DATE_TRUNC(...) ...
       shows results + summary

Build a dashboard

In Claude Code use /rawquery:dashboard or just describe what you want; in Codex, invoke $rq (or describe it directly — Codex picks the skill up from the description field):

You: "Build me a dashboard with monthly revenue, top customers, and churn rate"

Agent: explores your data
       writes + saves 3 queries
       creates 3 charts (line, horizontal_bar, number)
       assembles a page
       publishes it
       gives you the URL

Connect a new source

You: "Connect my production Postgres at db.example.com"

Agent: creates the connection
       tests it
       triggers initial sync
       describes the tables once synced

Explore unfamiliar data

In Claude Code, use /rawquery:explore after connecting a new source:

/rawquery:explore my_new_source

In Codex, invoke $rq then ask to explore the source — list tables, describe columns, suggest useful queries. The skill tells the agent how to drive rq for the same flow.