pre-alpha MIT MCP 1.x DuckDB node ≥ 20

Open-source onchain analytics, MCP-native.

Self-hosted, AI-agent-driven analogue of Dune Analytics. Same Parquet + dbt + curated tables on your laptop. 20 MCP tools so AI agents drive the analysis — not just humans clicking buttons.

Try in browser View on GitHub

What it is

MCP-native

14 tools spanning discovery, execution, semantics, memory, and output: list_tables, describe, query, metric, chart_render, report, recall, estimate_cost, budget_set, and more. Agents drive it. Humans optional.

Self-hosted

Parquet on local disk or S3. DuckDB by default; Trino and ClickHouse pluggable. No vendor API key. No usage cap. Your data never leaves your boundary.

Cost-aware

Pre-flight estimate_cost reads the query plan before execution. Per-session budget_set caps spend. Hard stop before a runaway query bills you.

Quickstart

# clone, install, seed
git clone https://github.com/Jacksstt/chainq.git ~/.chainq
cd ~/.chainq && pnpm install && pnpm seed
pnpm test

Install paths

Codespaces

One click. Hosted VS Code with chainq pre-seeded.

Open →

WASM playground

No install. DuckDB-WASM in your browser.

Launch →

Docker

docker compose -f docker/docker-compose.yml up

Compose file →

Fly.io / Render

One-click deploy templates. Public URL, free tier.

Deploy →

Local from source

pnpm install && pnpm seed && pnpm mcp:serve

Install guide →

Why

Dune was built when humans wrote SQL. In 2026, the primary consumer of onchain data is increasingly an AI agent. Agents need machine-readable schemas, cost estimates before execution, structured errors, and persistent memory of what they already queried.

Self-hosted teams need to keep sensitive data inside their boundary. chainq fills the gap: fully OSS, self-hosted, MCP-first, SQL-open.