Bring live roads to your assistant.
CommuteScout exposes its entire data layer as an MCP server, so Claude (or any MCP client) can answer questions about the roads in 37 states with live data instead of guesses.
https://mcp.commutescout.com/mcpLooking for the REST API, keys, limits and the full tool reference? See the developer documentation.
No key or account needed. It is the same server behind commutescout.com, with the same live feeds. Not running an MCP client? Every tool is also a plain HTTP GET at mcp.commutescout.com/v1, with an OpenAPI reference. Need more than the per-address limits? Sign in on the live map and create a key under Settings.
View the full reference on GitHubAdd the connector
Add a custom connector with the URL above (streamable HTTP), or run the server locally over stdio for Claude Desktop or Claude Code. No key or account needed either way.
Ask about a drive
Ask about a route or region in plain English. The assistant already knows 17 curated corridors, from I-80 Sacramento to Reno to the Bay Area freeways.
Get sourced answers
Every answer reads from the same live feeds behind the map, with per-source timestamps and context that changes the advice.
Ten tools, one live data spine
check_route(from_place, to_place, from_coords?, to_coords?)Everything active along a major corridor (17 curated corridors: I-80 Sacramento-Reno, US-50 to Tahoe, I-5, US-101, SR-17, SR-99, SR-1, I-15 to Vegas, Bay Area freeways, Tahoe locals), ordered by miles along the route
check_region(region)One-call report for a whole region (Bay Area, SoCal, Sierra, Central Valley, and four more): exact counts, incidents severity-sorted, full closures first, capped lists that say when they truncate
get_incidents(highway?, area?, center?, radius_km?)Live CHP incidents by route, dispatch area, or a point and radius
get_lane_closures(route?, district?, center?, radius_km?)Closures in place right now, classified per the closure taxonomy
get_chain_controls(route?, center?, radius_km?)Current chain requirements; says "none active" explicitly in the off-season
get_wildfires(near_route?, center?, radius_km?)Active fires with size, containment, and mapped perimeter edges, flagged near major highways
get_cameras(center?, route?, radius_km?, limit?)Roadside camera snapshots, each verified live before it is returned (offline placeholder frames are filtered by image freshness)
get_road_signs(route?, center?, radius_km?)What changeable message signs are displaying right now, verbatim
rank_routes(by?, limit?)All 17 corridors ranked by live events or measured congestion, with reasons; answers "what are the busiest routes right now"
get_nearby_events(center, radius_km?, kinds?)Live road events near a point across every covered state, 37 states, not just California; the fallback for locations outside California, near a state border, or when a California tool comes back empty
Route and region reports also carry context that changes the advice: weather alerts sampled along the trip, road-weather stations reporting something notable, recent significant earthquakes, the signs and cameras along the way, and live speeds when a TomTom key is set. Route names are normalized ("17", "hwy 50", "I80" all work), and docstrings are written for the LLM consumer: what the data is, its refresh cadence, and its limits.
Run it locally
The server is a single Python package with zero required keys. Config for Claude Desktop or Claude Code:
{
"mcpServers": {
"commutescout": {
"command": "uvx",
"args": ["--from", "git+https://github.com/nicglazkov/commutescout", "ca-roads-mcp"]
}
}
}From a checkout: pip install . then ca-roads-mcp for stdio, or ca-roads-mcp --transport http for streamable HTTP on $PORT. The http transport is tuned for Cloud Run (binds 0.0.0.0, host-header checks off), so bind it to localhost when running it on your machine: ca-roads-mcp --transport http --host 127.0.0.1.
How good are the answers?
An eval suite with recorded fixtures and 91 golden questions runs on demand, not on a release schedule. Each published scorecard names the version it reflects.
See the scorecard