YOUR FIRST MCP MISSION

Ask a question.
Follow the calculation.

Can a six-stop tour fit into twenty minutes? Your assistant asks. Rust connects two engines and an explicit conversion. You can inspect every handoff.

Try the same mission Connect your assistant
A questionWill this tour meet my deadline?
→
A checked planRoute → minutes → simulation
→
An inspectable resultNumbers, assumptions and limits
01
CONNECT ONCE

Give your assistant the two mission tools.

In an assistant that supports remote MCP, add this Streamable HTTP URL using its connection settings. Public access needs no API key. Host support and permissions depend on your assistant.

https://scorecompute.com/mcp?tools=compose_capabilities,execute_composition

The focused connection exposes the planner and executor. The ordinary endpoint still lets you call each engine directly. Connection instructions and access limits →

02
ASK WITH EXPLICIT ASSUMPTIONS

A request your assistant can reproduce.

Use ScoreCompute to assess this geometric tour. Points in kilometres: (0,0), (4,0), (5,3), (3,5), (0,4), (2,2). Start at the first point and return to it. Assume 60 km/h and round each leg up to a whole minute. Assess a 20-minute deadline with independent uniform duration variation of ±25%, 10,000 samples and seed 42. First call compose_capabilities with the route-risk request below and explain the three steps. If it is executable, call execute_composition with exactly the same request. Report the returned distance, duration assumptions, deadline probability and sampling interval. Do not treat geometric distances as roads or a successful execution as a safe deadline. Do not invent missing results.
The exact request shared by both tools
{
  "goals": [
    "simulate_plan.result.v1"
  ],
  "available": [
    "route_optimizer.request.v1",
    "route.speed_kmh.v1",
    "duration.independent_uniform.v1",
    "sampling.fixed.v1",
    "budget.minutes.v1"
  ],
  "constraints": {
    "allowed_tools": [
      "route_optimizer",
      "simulate_plan"
    ],
    "local_only": true,
    "max_steps": 3,
    "max_search_states": 128,
    "max_structural_cost": 32
  },
  "inputs": {
    "route_optimizer": {
      "metric": "euclidean_km",
      "points": [
        {
          "x": 0,
          "y": 0
        },
        {
          "x": 4,
          "y": 0
        },
        {
          "x": 5,
          "y": 3
        },
        {
          "x": 3,
          "y": 5
        },
        {
          "x": 0,
          "y": 4
        },
        {
          "x": 2,
          "y": 2
        }
      ],
      "return_to_start": true,
      "time_budget_ms": 1000
    },
    "simulate_plan": {
      "budget": 20,
      "uncertainty": 0.25,
      "samples": 10000,
      "seed": 42
    }
  },
  "adapters": {
    "route_durations.v1": {
      "speed_kmh": 60
    }
  }
}
Download arguments

ScoreCompute does not host a chat model here. Your assistant interprets the request and calls the tools. The plan and calculations run in Rust; unsupported paths are reported explicitly.

03
FOLLOW THE HANDOFFS

Two calls. Three calculated steps.

  1. compose_capabilities

    Inspect before calculating.

    The planner returns an admissible dependency graph. Planning does not run the numerical engines.

  2. execute_composition

    Run the approved chain.

    Rust checks the request again, orders the points, converts each segment with your stated speed, then evaluates the durations under your uncertainty model.

  3. Read the returned evidence

    Explain what the numbers establish.

    Keep the route, rounded durations, sampling interval and limits. “Succeeded” means the calculation completed; it does not mean the deadline is likely to be met.

See the engines work together
04
A RECORDED REAL EXECUTION

Inspect an example before your own run.

This is a saved execution of the exact request above through a private MCP client, using synthetic points. Loading it performs no calculation. It is an execution record, not a signed Verify certificate or an independent scientific audit.

ALSO REPRODUCIBLE WITHOUT A CHAT APP

Use the same MCP calls from a terminal.

Download the arguments above as first-mission.arguments.json. With Node.js 20 or later, run the planning command. Inspect its output, then run execution if it matches the approved route.

Linux / macOS commands
cat first-mission.arguments.json | npx --yes --package https://scorecompute.com/downloads/scorecompute-call-0.14.1.tgz scorecompute-call compose_capabilities --url 'https://scorecompute.com/mcp?tools=compose_capabilities,execute_composition' --stdin

cat first-mission.arguments.json | npx --yes --package https://scorecompute.com/downloads/scorecompute-call-0.14.1.tgz scorecompute-call execute_composition --url 'https://scorecompute.com/mcp?tools=compose_capabilities,execute_composition' --stdin
Windows PowerShell commands
Get-Content -Raw first-mission.arguments.json | npx --yes --package https://scorecompute.com/downloads/scorecompute-call-0.14.1.tgz scorecompute-call compose_capabilities --url 'https://scorecompute.com/mcp?tools=compose_capabilities,execute_composition' --stdin

Get-Content -Raw first-mission.arguments.json | npx --yes --package https://scorecompute.com/downloads/scorecompute-call-0.14.1.tgz scorecompute-call execute_composition --url 'https://scorecompute.com/mcp?tools=compose_capabilities,execute_composition' --stdin

npx downloads and runs the ScoreCompute client. The computations run on the remote service. Each execution is subject to the same public limits; no automatic retries are included. Client download and checksum →