RouteKit MCP
Route optimization API for Brazil with tools for geocoding, distance matrices, isochrones, and advanced vehicle routing, supporting constraints like time windows, skills, and multi-profile fleets.
README
RouteKit MCP
<!-- mcp-name: io.github.luisbretasjr/routekit -->
Route optimization API for Brazil. First MCP server for route optimization.
Install
npx routekit-mcp
Or configure in your MCP client:
Claude Desktop / Claude Code
{
"mcpServers": {
"routekit": {
"command": "npx",
"args": ["-y", "routekit-mcp"],
"env": {
"ROUTEKIT_API_KEY": "YOUR_KEY"
}
}
}
}
Direct HTTP (no install needed)
{
"mcpServers": {
"routekit": {
"type": "url",
"url": "https://routekit.nexterait.com.br/mcp",
"headers": {
"X-API-Key": "YOUR_KEY"
}
}
}
}
Get API Key
Get a free key (50 calls/month) at routekit.nexterait.com.br/static/signup.html
Tools
- calculate_route -- Driving route between points with real road distances and ETAs. Supports waypoints, encoded polyline geometry, turn-by-turn steps, and up to 3 alternative routes.
- distance_matrix -- NxN distance/duration matrix between up to 5000 locations (transparently chunked for large sets, cached for repeated queries).
- snap_to_road -- Snap a raw GPS coordinate to the nearest road segment. Returns the snapped point, distance from the original, and road name.
- isochrone -- Area reachable from a point within N minutes by car. Returns a GeoJSON Polygon useful for coverage analysis, depot placement, and service area definition.
- geocode -- Convert a Brazilian address or place name to latitude/longitude. Restricted to Brazil, in-house geocoder with no third-party dependency. Built-in fuzzy fallback handles abbreviations ("Av" -> "Avenida"), accents, and token drops.
- geocode_batch -- Geocode up to 50 addresses in a single call with parallel execution. Perfect for delivery manifests or client lists.
- reverse_geocode -- Convert lat/lon to a human-readable Brazilian address with components (road, neighborhood, city, state, postcode).
- compare_routes -- Run
optimize_routesunder multiple balance strategies (minimize_vehicles, balance_tasks, minimize_distance) and return a side-by-side comparison with winners per metric. - validate_eta -- Simulate a pre-defined route in Python and return the real arrival times plus any constraint violations. Use when you already have a fixed order of stops and just need the ETAs.
- optimize_routes -- Advanced VRP solver:
tasks(single-location jobs) and/orshipments(paired pickup+delivery, same vehicle)- Constraints: time windows (multiple per task), skills, multi-dimensional capacity, breaks, priorities
- Per-vehicle costs (
fixed,per_hour,per_km,per_task_hour),max_tasks,max_travel_time_min,max_distance_km balance_mode:minimize_vehicles(default),balance_tasks, orminimize_distancesetup_minseparate fromservice_min(setup charged once per location)setup_per_type/service_per_typefor different times per technician/vehicle typevehicle.typetags for per-type time/cost modelingend_lat/end_lonfor open routes with different destination (e.g. driver ends at home)- Mixed-profile fleets (car / bike / foot) with per-vehicle
profile include_geometry=truereturns encoded polylines per route (for maps)matrix_overrides: user-provided per-pair travel time / distance overrides (for toll routes, restricted zones, known detours)- Up to 500 tasks / 250 shipments / 100 vehicles per request
- Route-level and step-level
violationsreported when constraints are tight
Example
Ask your AI assistant:
"Geocode these addresses and optimize 10 deliveries in Sao Paulo for 2 drivers -- the senior finishes each stop in 20 min, the junior in 45 min. Balance tasks evenly, end both at the same warehouse, include the route polylines so I can draw them on a map."
The AI will call geocode, then optimize_routes with vehicle.type, service_per_type, balance_mode, end_lat/end_lon and include_geometry -- no manual coordinate entry needed.
What's new in 1.4.1
- Per-tool analytics: the dashboard now shows a breakdown of which tools your API key has been calling (horizontal bars, ordered by frequency).
- 80% limit warning: when your usage crosses 80% of the monthly limit, the dashboard shows a yellow banner reminding you to upgrade before new calls are rejected.
/billing/usagealso exposesthreshold_80andthreshold_100flags. - Welcome email: new customers get an onboarding email right after OTP verification, with their API key, ready-to-paste Claude Desktop config, a curl snippet and doc links.
What's new in 1.4.0
calculate_routeturn-by-turn: passinclude_steps=trueto get a list of maneuvers with street names, distances and durations per step.calculate_routealternatives: passalternatives=1..3to get alternate routes (useful for "fastest vs shortest vs no-toll" comparisons).- Fuzzy geocoding:
geocodenow silently retries with abbreviation expansion ("Av" -> "Avenida"), accent normalization and token-drop heuristics when the exact query returns nothing. The response includesvariant_usedshowing which transformation matched. Setfuzzy=falseto disable. - REST API: the same 10 tools are now available via plain HTTP POST under
https://routekit.nexterait.com.br/v1/<tool_name>. Same X-API-Key auth, same metering. The publicGET /v1/lists all endpoints with their descriptions. Useful for integrations that don't speak MCP. - Billing portal return flow: after managing a subscription on Stripe, the user lands back on the dashboard with a success banner instead of a raw JSON page.
What's new in 1.3.0
compare_routestool: compare the optimizer under multiple strategies and get winners per metric (distance, travel time, vehicles used). Runs strategies in parallel.validate_etatool: simulate a pre-defined route and get real ETAs plus violations without re-optimizing. The sequence you provide is strictly respected.matrix_overrideson optimize_routes: patch specific pairs in the distance/time matrix with your own values (toll routes, ferries, known detours) without forcing users to send a whole NxN matrix.
What's new in 1.2.3
geocode_batch: new tool accepting up to 50 addresses per call. Internally runs queries in parallel with bounded concurrency.- Mixed-profile fleets:
optimize_routesnow fetches separate distance/duration matrices per vehicle profile (car/bike/foot). When a profile container isn't deployed, it falls back to car routing transparently. - Smarter image check: the weekly OSRM image check now queries registry tags and ranks stable semver releases (filtering debug/assertions/rc variants).
What's new in 1.2.2
- Matrix cache: repeated distance/duration matrix queries hit a local cache and return ~5x faster. Transparent to callers.
- Automatic chunking: distance_matrix now handles up to 5000 locations by splitting into parallel sub-block queries.
- Retry + backoff: all upstream calls now retry transient failures (connect errors, 5xx) with exponential backoff + jitter before surfacing an error.
- Cleaner error messages: "temporarily unavailable" vs "no result found" so the LLM can decide whether to retry.
What's new in 1.2.0
- Geocoding: new
geocodeandreverse_geocodetools for Brazil. Users can speak in addresses instead of coordinates. No third-party API dependency. - Snap-to-road: new
snap_to_roadtool to clean up noisy GPS before routing. - Isochrone: new
isochronetool to compute areas reachable in N minutes. - Technician modeling:
vehicle.type,setup_per_type,service_per_type,costs.per_task_hourfor per-type time and cost realism. - Open routes:
end_lat/end_lonper vehicle for drivers that end at a different place than they started. - Multi-profile fleets: label vehicles with
profile(car/bike/foot/truck).
What's new in 1.1.0
shipments(paired pickup+delivery) supportbalance_modeto distribute tasks evenly across a fleetinclude_geometryto return route polylines from the optimizer- Route
violationsnow surfaced in the response for diagnosis setup_minvsservice_minsplit for realistic location costs- Matrix limit raised from 200 to 2000 locations
- Advanced VRP engine upgrade:
per_kmcosts andmax_distance_kmconstraints now active
Pricing
| Tier | Calls/Month | Price |
|---|---|---|
| Free | 50 | R$ 0 |
| Starter | 1,000 | R$ 500/month |
| Professional | 10,000 | R$ 3,000/month |
Links
Provider
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。