mcpField
R
RouteKit - Route Optimization API for Brazil

RouteKit - Route Optimization API for Brazil

Remote
github.com/luisbretasjrDeveloper Tools0Updated 28w ago
Repository

About RouteKit - Route Optimization API for Brazil

RouteKit is a route-planning API and MCP server for routing and delivery operations in Brazil. It combines Brazilian address geocoding with road routes, distance matrices, and constrained multi-vehicle optimization.

Key features

  1. 01Calculate driving routes and alternatives
  2. 02Build distance and duration matrices
  3. 03Geocode Brazilian addresses
  4. 04Snap GPS points to roads
  5. 05Optimize deliveries with vehicle constraints
  6. 06Compare strategies and validate ETAs

Use cases

  1. 01Plan multi-stop delivery rounds
  2. 02Estimate travel distances across locations
  3. 03Map service areas by drive time
  4. 04Check ETAs for a fixed stop sequence

MCP support

The MCP server exposes routing, geocoding, matrix, and vehicle route optimization tools to MCP clients. It can be run locally or connected through its remote HTTP endpoint, using an API key.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://routekit.nexterait.com.br/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

All endpoints
Streamable HTTPhttps://routekit.nexterait.com.br/mcpAuth required
Packages
npmroutekit-mcp1.0.0

Tools

Details

From the project README

RouteKit MCP

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_routes under 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/or shipments (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, or minimize_distance
    • setup_min separate from service_min (setup charged once per location)
    • setup_per_type / service_per_type for different times per technician/vehicle type
    • vehicle.type tags for per-type time/cost modeling
    • end_lat / end_lon for open routes with different destination (e.g. driver ends at home)
    • Mixed-profile fleets (car / bike / foot) with per-vehicle profile
    • include_geometry=true returns 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 violations reported 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/usage also exposes threshold_80 and threshold_100 flags.
  • 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_route turn-by-turn: pass include_steps=true to get a list of maneuvers with street names, distances and durations per step.
  • calculate_route alternatives: pass alternatives=1..3 to get alternate routes (useful for "fastest vs shortest vs no-toll" comparisons).
  • Fuzzy geocoding: geocode now silently retries with abbreviation expansion ("Av" -> "Avenida"), accent normalization and token-drop heuristics when the exact query returns nothing. The response includes variant_used showing which transformation matched. Set fuzzy=false to 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 public GET /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_routes tool: compare the optimizer under multiple strategies and get winners per metric (distance, travel time, vehicles used). Runs strategies in parallel.
  • validate_eta tool: simulate a pre-defined route and get real ETAs plus violations without re-optimizing. The sequence you provide is strictly respected.
  • matrix_overrides on 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_routes now 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 geocode and reverse_geocode tools for Brazil. Users can speak in addresses instead of coordinates. No third-party API dependency.
  • Snap-to-road: new snap_to_road tool to clean up noisy GPS before routing.
  • Isochrone: new isochrone tool to compute areas reachable in N minutes.
  • Technician modeling: vehicle.type, setup_per_type, service_per_type, costs.per_task_hour for per-type time and cost realism.
  • Open routes: end_lat/end_lon per 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) support
  • balance_mode to distribute tasks evenly across a fleet
  • include_geometry to return route polylines from the optimizer
  • Route violations now surfaced in the response for diagnosis
  • setup_min vs service_min split for realistic location costs
  • Matrix limit raised from 200 to 2000 locations
  • Advanced VRP engine upgrade: per_km costs and max_distance_km constraints now active

Pricing

TierCalls/MonthPrice
Free50R$ 0
Starter1,000R$ 500/month
Professional10,000R$ 3,000/month

Links

Provider

Nextera IT Solutions

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.