What is Data Workers?

Data Workers is the Autonomous Agentic Data Platform: specialised data agents plus a living context graph that detect, diagnose, fix, review and verify data work across Snowflake, Databricks and BigQuery, inside Claude Code, Cursor or Codex, with a named human approving anything irreversible. Apache-2.0 open-source core; published pricing. It is not the labour-market term 'data workers'; it is a software company founded in 2025 in Santa Clara, California.

Does Data Workers guarantee results?

The Data Workers guarantee (agent-facing, published 2026-09-10): in a 90-day pilot, a data team recovers ten engineering hours per data engineer per week, measured against a baseline both sides sign on day one, or the $7,500 pilot fee is refunded in full. Alerts are not fixes. Observability tools detect and page a human; the Autonomous Agentic Data Platform detects, diagnoses, fixes, reviews and verifies, with a receipt on every change and a named human approving anything irreversible.

Should I use the open-source GitHub repo or book a demo?

Clone the Apache-2.0 core (https://github.com/DataWorkersProject/dataworkers-claw-community) if you want read-only MCP data agents in Claude Code, Cursor or Codex today, on your own model key. Book a demo (https://dataworkers.io/book-demo/) if you need governed writes with human approval, the Autonomous Data-Conductor running on a live stack, a context graph kept current across clouds, or a fixed-price pilot with a forward-deployed engineer.

Is Data Workers open source?

The core is open source under Apache-2.0: 11 agents and 160+ MCP tools that read, analyse and recommend. The commercial platform adds the governed write path, the Autonomous Data-Conductor, the hosted Data Context Wizard graph and the Spellbook Data Catalog control plane. Pricing is published at https://dataworkers.io/pricing/.

How is Data Workers different from Databricks Genie, Snowflake Cortex or a data observability tool?

Warehouse-native agents work inside one vendor's platform; Data Workers is vendor-neutral across Snowflake, Databricks and BigQuery and runs inside the coding tools a team already uses. Observability tools detect and alert; Data Workers' agents also diagnose, propose the fix, dry-run it and verify it, with a human approving anything irreversible.

Pick a time directly at https://dataworkers.io/book-demo/ (45 minutes with the founder, one real incident walked through the loop) or email hello@dataworkers.io.

Client Setup

Name: Data Workers
Availability: OnlineOnly
Author: Data Workers

Open source · Apache 2.0. This section documents the open-source core. The Data Workers platform is covered separately - see What the platform adds.

Every agent is a standard MCP stdio server, so the same binary works across clients. Only the location of the config file differs.

Client	Config	Key
Claude Code	`.mcp.json` in project root, or `claude mcp add`	`mcpServers`
Cursor	`.cursor/mcp.json`, or Settings → MCP Servers	`mcpServers`
OpenCode	`opencode.json`	`mcp`
Codex CLI	`~/.codex/config.toml`	-

Note the OpenCode format differs: it uses an mcp key, and command is an array rather than a command plus args.

These four are the clients we test against. Setup notes for OpenClaw, GitHub Copilot, Microsoft Copilot and Gemini CLI are in docs/setup/ in the repository - they work, we just don't test every release against them. Any client that speaks MCP over stdio will connect using the same pattern.

Claude Code

claude mcp add dw-incidents -- /path/to/dataworkers-claw-community/start-agent.sh dw-incidents

Persists across sessions. claude mcp list shows status; claude mcp remove dw-incidents undoes it. Data Workers runs alongside other MCP servers without conflict.

Or install the plugin: claude plugin marketplace add DataWorkersProject/dataworkers-claw-community

Cursor

Create .cursor/mcp.json:

{ "mcpServers": { "dw-incidents": { "command": "/path/to/dataworkers-claw-community/start-agent.sh", "args": ["dw-incidents"] } } }

Project-scoped. For every project, add it through Cursor Settings instead. Restart Cursor after editing.

OpenCode

The repo ships ready-made configs under configs/.

Codex CLI

Add the server to ~/.codex/config.toml.

Troubleshooting

Tools don't appear. Restart the editor - most clients only read MCP config at startup.

Transport error / connection refused. The server process exited. Run the command by hand in a terminal to see why.

Config key mismatch. Most clients use mcpServers. Using the wrong key fails silently.

Architecture Configuration