What it does
Gigalytics MCP is a thin agent in front of your dedicated Gigalytics MCP server — the one your team already uses internally to query aggregate FSD telemetry. Once the server is connected to Habitats once at the org level, the agent can pull a release window, split it by road class, and render a chart inline in any Habitat it's attached to.
Nothing about your data leaves the MCP boundary. The agent only sees the rolled-up rows the server returns from drives.window(); it never touches raw drive logs. Charts are rendered locally on the agent host before they're attached to the workstream as PNG.
Capabilities
- ✓ Release comparison. Pulls no-intervention rate, miles-per-intervention, and the highway/surface split for any release window. Scoped to the canonical drive set.
- ✓ Daily curves. Breaks an N-day window into a daily series so you can spot rollout outliers (day-1 spikes, post-update regressions) at a glance.
- ✓ Local chart rendering. Histograms and line charts render on the agent host, attach to the workstream as PNG, and stay viewable inside the artifact drawer.
- ✓ Read-only by design. Never writes back to the MCP server. No cache, no mutation; every run is a fresh pull.
- ✓ Token-scoped. Auth is per-org, never per-Habitat, so revoking the token at the Gigalytics side cleanly disconnects every Habitat at once.
Connection
This agent talks to your Gigalytics server over MCP (Model Context Protocol). Once the server is registered as a connection on your Habitats account, the agent can call any of its tools from any Habitat it's attached to.
| Tool | Surface | Description |
|---|---|---|
health() |
read | Server liveness check; returns version + drive-set freshness. |
drives.window(version, days) |
read | Aggregate FSD telemetry for a release version over an N-day window. |
drives.split(version, by) |
read | Split a window by road class (highway/surface), time of day, or weekday. |
compute.histogram(by, metric) |
compute | Render a histogram from drive metadata; agent attaches the rendered PNG to the workstream. |
Configure
Paste this server config into Settings → Connections → Add MCP, or run the CLI:
# Register the Gigalytics MCP server once at the org level. $ habitats connections add \ --name "Gigalytics MCP" \ --type mcp \ --endpoint "mcp.gigalytics.internal" \ --token "$GIGALYTICS_ORG_TOKEN" # Then attach the agent to a Habitat. $ habitats agents attach gigalytics-mcp --habitat gigalytics
Once connected, this agent appears in the @-mention picker of every Habitat you attach it to. Token rotation is handled at the Gigalytics side; Habitats reuses whatever the server accepts.
Changelog
Maintained by the Gigalytics team — versioning is independent of your Habitat — pinning is supported per Habitat in Configure.
drives.split for road-class breakdowns. Charts now include the daily curve over the N-day window so rollout outliers (day-1 spikes, post-update regressions) are visible without a second query.health() on every run and surfaces freshness in the workstream.compute.histogram and the no-intervention-rate aggregator. Median miles-per-intervention is now computed across complete trips ≥ 5 miles to filter out parked-car anomalies.health() + drives.window() only. No auto-charting; agent dropped raw rows into the workstream as a CSV artifact.