sparrowflight · mcp
The dataset speaks MCP.
So the AI analyst comes built in.
So the AI analyst comes built in.
Model Context Protocol is how agents reach tools and data. sparrowMCP
puts an MCP server in front of any Arrow Flight endpoint: Claude — or any MCP-capable
agent — discovers the catalog, pulls series, and runs analytics natively. No CSV
uploads, no copy-paste, no retrieval contortions.
Database MCP servers are SQL bridges. sparrowMCP brings an
econometrics toolbox with it — and the analysis runs
server-side, at the data. The agent's context window sees results, not megabytes
of rows.
Discover
Catalog tools — search millions of series by name, unit, frequency, source; browse the category tree
Data
Series pulls over Arrow Flight — columnar on the wire, right-sized summaries in the context window
Analytics
41 analytics tools — ARIMA / Prophet / Theta ensembles, correlation sweeps, cointegration, VAR, breakpoints, seasonality, outliers — all computed server-side
Narrative
Signals → contradictions → regime — scored signals roll into named headlines; report runs are saved and diffed, so the agent knows what changed since yesterday
Reach
Any Flight server — and Flight SQL engines (Dremio, InfluxDB 3) over the same wire
one question, whole pipeline
"what moved against the WTI crack spread this quarter?"
→ search · get_data · rolling_correlation · chart → ranked answers, computed at the server — the data never leaves.
→ search · get_data · rolling_correlation · chart → ranked answers, computed at the server — the data never leaves.
The proof is in production: energyscope-mcp ships on PyPI —
62 tools over 2.8 million series and
136 million rows. It researches and writes the
EnergyScope morning oil brief, and its signal layer once identified a market regime
five days before the cause was publicly known. The core
underneath is domain-blind: every vertical built on Sparrow gets its analyst on
day one.
sparrowflight.io · 2026