DataRaum
- 3 repo stars
- Updated July 23, 2026
- dataraum/dataraum
io.github.dataraum/dataraum is a MCP server that supplies pre-computed metadata context for AI-driven data analytics over your workspaces.
About
DataRaum is an MCP server that front-loads semantic and structural context about your analytics estate so agents answer questions against the right tables, metrics, and business definitions instead of guessing from bare DDL. developers shipping dashboards or internal data products often stall because metadata lives in people's heads or scattered docs; DataRaum materializes that context in a workspace-oriented home directory and exposes it through stdio for Claude Code and peers. The server leans on Anthropic for semantic analysis, so you trade an API key for richer column descriptions, entity relationships, and exportable session artifacts. Version 0.2.1 is early but the positioning is clear: analytics and exploration, not ETL authoring from scratch. Install via uvx, set DATARAUM_HOME on a persistent volume, and wire the MCP entry in your agent config. Best when you already have tables or warehouses and need trustworthy context; less ideal if you only need a one-off CSV plot with no ongoing metadata discipline.
- Pre-computed metadata context engine aimed at AI-driven analytics workflows
- PyPI package dataraum 0.2.1 runnable via uvx stdio MCP transport
- ANTHROPIC_API_KEY required for LLM-powered semantic analysis over metadata
- DATARAUM_HOME configures root for workspaces, sessions, and exports
- Open-source GitHub repo dataraum/dataraum
DataRaum by the numbers
- Data as of Jul 24, 2026 (Skillselion catalog sync)
claude mcp add --env ANTHROPIC_API_KEY=YOUR_ANTHROPIC_API_KEY --env DATARAUM_HOME=YOUR_DATARAUM_HOME dataraum -- uvx dataraumAdd your badge
Show developers this MCP server is listed on Skillselion. Paste this into your README.
| repo stars | ★ 3 |
|---|---|
| Package | dataraum |
| Transport | STDIO |
| Auth | Required |
| Last updated | July 23, 2026 |
| Repository | dataraum/dataraum ↗ |
What it does
Give your agent pre-computed dataset and column metadata so analytics questions need less raw SQL spelunking and fewer wrong joins.
Who is it for?
Best when you have an existing warehouse or analytics stack and want cheaper, safer agent analytics sessions.
Skip if: Skip if you have no Anthropic key budget or no appetite to maintain DATARAUM_HOME workspaces and exports.
What you get
After registration, queries and plans can lean on DataRaum’s semantic metadata layer and exported session context instead of rediscovering schema every time.
- Agent-addressable pre-computed metadata context for datasets
- Workspace-scoped sessions and exports under DATARAUM_HOME
- Semantic-enriched analytics grounding for downstream agent queries
By the numbers
- Published MCP server version 0.2.1
- PyPI identifier dataraum with stdio transport
- ANTHROPIC_API_KEY marked required in server manifest
README.md
DataRaum
A rich metadata context engine for AI-driven data analytics.
Traditional semantic layers tell BI tools "what things are called." DataRaum tells AI "what the data means, how it behaves, how it relates, and what you can compute from it."
Monorepo layout
packages/
├── engine/ # Python — pipeline, detectors, Starlette kernel shell
├── cockpit/ # TypeScript — TanStack Start web UI
└── infra/ # docker-compose orchestration
Each package has its own README. Start there if you're working in a specific package.
Status — transitioning to v1
DataRaum is mid-pivot. v0.2.x exposed a 12-tool MCP server over HTTP. That transport is gone. v1 is a 3-verb kernel + cockpit:
- engine — Python pipeline + Starlette kernel shell exposing three verbs over HTTP:
/measure(SSE),/query(Arrow),/probe(read-only SQL), plus/health. - cockpit — TanStack Start app that hosts the chat surface, renders the agentic widgets, and reads the engine's metadata schema directly via Drizzle (introspected per workspace).
Today the substrate boots and you can poke /health. The 3 kernel verbs are 501 stubs and get filled in phase-by-phase per the DAT-339 pivot. No end-user surface yet — if you need v0.2.x MCP behavior, pin dataraum==0.2.2.
Quick start
# Set the LLM key
cp packages/infra/.env.example packages/infra/.env
echo "ANTHROPIC_API_KEY=sk-ant-..." >> packages/infra/.env
# Bring up Postgres + control plane + cockpit
docker compose -f packages/infra/docker-compose.yml up -d --wait
# Verify the substrate
curl -fsS http://localhost:8000/health
# Open the cockpit
open http://localhost:3000
For UI iteration, run the cockpit dev server outside docker for hot reload — see packages/cockpit/README.md.
Develop
- Engine (Python):
cd packages/engine && uv sync --group dev && uv run pytest --testmon tests/unit -q. Seepackages/engine/README.mdandpackages/engine/CLAUDE.md. - Cockpit (TypeScript):
cd packages/cockpit && bun install && bun run dev. Seepackages/cockpit/README.mdandpackages/cockpit/CLAUDE.md. - Pull the engine metadata schema (cockpit):
cd packages/cockpit && DATARAUM_WORKSPACE_ID=<id> METADATA_DATABASE_URL=<url> bun run db:pull:metadata. Re-run after the engine adds/changes SQLAlchemy models.
Documentation
User-facing docs live in packages/engine/docs/ and are published via Zensical.
License
MIT — see LICENSE.
Recommended MCP Servers
1ClickReport40 tools for marketing analytics, campaign management, and AI monitoring across 6 platforms.
222wcnm BilistalkermcpTrack Bilibili creators and get the latest updates on videos, dynamics, and articles.7
ABMeterFeature flagging and A/B testing platform with AI-first experimentation workflows.
Abs McpQuery the ABS in plain English — CPI, labour force, building approvals, migration, GDP.
AdAdvisor MCP ServerQuery Meta Ads performance data — accounts, campaigns, ad sets, ads, metrics & settings.
ADCIRC Model Setup MCP ServerADCIRC model configuration debugging, parameter lookup, and validation5How it compares
Metadata context engine MCP, not a replacement for dbt, Airbyte, or a live SQL execution MCP.
FAQ
Who is io.github.dataraum/dataraum for?
Developers and analytics-minded founders who use AI agents to explore metrics and need grounded table and column context.
When should I use io.github.dataraum/dataraum?
Use it in Grow when you are iterating on dashboards, funnels, or ad hoc analysis and want the agent to reuse pre-computed semantic metadata.
How do I add io.github.dataraum/dataraum to my agent?
Run the dataraum PyPI package via uvx as stdio MCP, set ANTHROPIC_API_KEY and DATARAUM_HOME, then add the server block to Claude Code, Cursor, or Codex MCP settings.