Charge Atlas MCP Server

Charge Atlas — global EV charging station data

Charge Atlas answers the question "where can I charge an electric vehicle here?" anywhere in the world. It exposes one consistent station shape and one consistent filter set across two coverage regions:

• coverage: "us_canada" — the United States (50 states + DC + territories) and Canada. Authoritative, refreshed daily, includes private and planned sites alongside public ones.
• coverage: "global" — every other country. Community-edited, strongest in the EU and UK, sparser in some emerging markets.

The right region is picked automatically from the search coordinates or country code; the AI never has to choose. The user-visible coverage field on every response simply labels which region produced the result so callers can warn appropriately (community-edited listings can be stale).

Interactive Map App

An interactive map dashboard is available alongside this chat. When the app panel is open, search results are automatically forwarded to the map — do NOT re-open the app. Just call the search and detail capabilities and the map will update with stations rendered as colored circle markers (green = DC Fast, orange = Level 2, gray = Level 1), click-through to station detail, and a panel list showing the top results.

Data Model

• Station — A single physical charging location with one street address. Each station carries one or more EVSE ports (Electric Vehicle Supply Equipment) and one or more connector types. The same parking lot is one station even if it has 12 stalls.
• Connector / Port — The physical plug. Common types: CCS Combo (most non-Tesla DC fast), CHAdeMO (older Japanese DC fast, Nissan Leaf), NACS / Tesla (Tesla Supercharger; growing US standard), J1772 (Level 2 AC; nearly all non-Tesla US/Canada AC), Type 2 Mennekes (European AC standard), Tesla Destination (Tesla-branded Level 2).
• Charge Level — L1 (120V AC, ~5 mi/hr — wall outlet), L2 (208–240V AC, ~25 mi/hr — most public AC), DC (DC fast, 50–350 kW, 100–250 mi in 30 min).
• Network — Operator brand (Tesla Supercharger, Electrify America, EVgo, ChargePoint, IONITY, Ionna, BP Pulse, Shell Recharge, Pod Point, etc.). Many stations are non-networked (e.g. workplace, hotel).
• Status — available, in_use, out_of_order, planned, temporarily_unavailable. Real-time availability is partial: only some networks publish it, and freshness varies.

Coverage Regions

Searches by latitude/longitude or bounding box automatically pick the right region from the coordinates. Searches that take a country code respect that code directly. Route-corridor search is US/Canada-only — for international routes, use the journey-planner capability instead.

Key Patterns

• Coordinates first: every search accepts latitude + longitude. Bounding-box search takes min_lat / min_lon / max_lat / max_lon (south-west and north-east corners). Distances are in statute miles unless specified.
• Filter by charger speed with min_charge_level = L1 | L2 | DC. To find road-trip-capable stations, pass DC.
• Filter by network by passing the network slug (e.g. Tesla, ChargePoint, EVgo, Electrify America, IONITY). Use the network listing capability to discover valid slugs for a country.
• Free-text location search is global. Place names, neighborhoods, addresses, and postcodes are geocoded automatically. Pass city + country on the first query ("central London, UK", "Madrid, Spain") rather than retrying with rephrased spelling. When the host supplies a country hint, pass it as country_bias.
• Station IDs are opaque tokens returned on every search row's id field. Pass them straight back into the detail capability — they round-trip unchanged. Detail responses for community-coverage stations also include up to five recent user reviews and check-in reports when available.
• Vehicle compatibility filtering: pass vehicle (e.g. tesla-model-3, nissan-leaf, kia-ev6) on any search to drop chargers whose connector set is incompatible with that car. The vehicle catalog capability returns every supported model slug.
• Local operator mix (city or neighborhood): When the user wants chargers on the map and how many belong to each operator in that area, one free-text location search with min_charge_level=DC is enough. The response includes network_breakdown (operator, count, share percentage) across the full matched set — read that for the tally and the station list for map detail in the same turn.
• Country-wide operator rankings: the country-summary capability returns pre-aggregated top_networks — use it for whole-country questions, not for a named neighborhood or city center.
• Head-to-head named operators in a country: the network-comparison capability ranks two-to-eight named networks by station count and share — useful for buy-decision questions like "IONITY vs Tesla in Spain".
• Journey planning: the journey planner takes free-text origin and destination, geocodes both, fetches a driving polyline, and returns DC fast chargers along the route. The polyline is tagged route_kind: "driving" for a real road route or "approximate" for a straight-line fallback when no road route is available — surface that to the user when it's approximate.
• Locale-aware units: search responses include a display_units hint derived from the host's BCP-47 locale (US/UK/Liberia/Myanmar → miles, everywhere else → km). All distances are also returned in both miles and km on every station so the renderer can pick the right one.
• Auto-locate: when latitude/longitude are omitted, ChatGPT clients pass the coarse user location via _meta.openai/userLocation and Charge Atlas uses it to default the search center. Country-summary and free-text location searches similarly default the country bias from the same hint.
• Result sizing: For map-first city or neighborhood queries, keep max_results at 50–75 unless the user explicitly needs an exhaustive inventory. The count field reports how many stations matched; network_breakdown summarizes operators across the full returned set even when the station table preview truncates.
• One response, two answers: When count and network_breakdown are already on a location search response, that single result answers both the map and the operator tally — no need to repeat the same place with alternate spelling before answering.
• Open-now: Charge Atlas only marks a station open_now: true when the hours string clearly says 24/7. Anything else is reported as unknown — half-broken hours parsing is worse than honest "we don't know".

Gotchas

• Real-time availability is sparse. Many networks (especially private/workplace chargers and the long tail of small operators) do not publish live status. Treat results without a status field as "operational per last directory update", not "free right now". Always check the network's own app before driving to a single specific station.
• Connector compatibility is critical. A Tesla Model 3 with a J1772 adapter can use J1772 and Tesla destination; a Nissan Leaf needs CHAdeMO for DC fast. Returning a station whose only DC port is CCS to a CHAdeMO car is a wrong answer. When the user names their car, narrow connector_types accordingly.
• NACS transition is mid-flight. As of 2024–2026 Tesla Supercharger sites are opening to non-Tesla cars on a rolling basis. Open access is published per-station — not all Superchargers in a city are open to non-Tesla EVs at the same time.
• DC kW ratings are a maximum, not a guarantee. A "350 kW" stall delivers that to a vehicle that can accept it; many EVs cap at 100–150 kW. Do not promise "30 minutes to 80%" without knowing the car's peak DC rate.
• International journey planning is sampled, not exhaustive. For long international drives the planner samples several waypoints along the polyline and runs a nearby search at each — adequate for trip planning, but may miss chargers between sample points on extremely long routes.
• coverage: "global" listings can be stale because the underlying directory is community-edited. A station marked operational there may have been decommissioned; cross-reference with the network's own site for high-stakes decisions.
• Distance units: every station carries both distance_miles and distance_km. The response also includes a display_units hint chosen from the user's locale — render whichever the hint asks for. Pass units: "miles" or units: "km" to override.
• Network breakdown scope: Country-summary and network-comparison answer country-scale questions. A query scoped to a neighborhood ("central London", "SoMa", "Kreuzberg") needs a location search — substituting country-level stats mis-answers the question.

Disclaimers

Coverage, real-time status, and network coverage vary by country and by network. This data is best-effort and not a substitute for the network's own routing app for time-critical or safety-critical journeys (long-range road trips, single-shot drives in unfamiliar territory). Pricing, plug-and-charge support, and amenity details (open hours, restrooms, food) are not guaranteed and should be confirmed at the network's own site.