Enhance tool description with optimization triggers and examples

[wspringer]

Summary

Enhance the HiGHS MCP tool description to include optimization trigger phrases, problem examples, and capability information to improve LLM discovery and usage.

Problem

The current tool description is minimal:

description: "Solve linear programming (LP) or mixed-integer programming (MIP) problems using HiGHS solver"

This makes it difficult for LLMs to recognize when the tool is relevant for user queries about optimization.

Proposed Solution

Update the tool description to include rich metadata within the MCP-compliant description field:

{
  name: "optimize-mip-lp-tool",
  description: `Solve optimization problems: minimize costs, maximize profits, allocate resources, schedule tasks, find optimal routes, and make the best assignments under constraints.

TRIGGERS: "find the best", "optimal solution", "minimize total", "maximize profit", "subject to constraints", "allocate resources", "scheduling problem"

PROBLEM TYPES:
• Transportation: Ship goods from warehouses to stores at minimum cost
• Production: Decide how much of each product to make with limited resources  
• Scheduling: Assign workers to shifts to meet demand
• Portfolio: Allocate investment across assets to maximize return
• Assignment: Match resources to tasks optimally
• Network Flow: Optimize flow through networks

CAPABILITIES:
• Problem types: Linear Programming (LP), Mixed-Integer Programming (MIP)
• Scale: Up to 1,000,000 variables and constraints
• Variable types: Continuous, integer, binary
• Algorithms: Simplex, interior-point, branch-and-bound
• Matrix formats: Dense and sparse (COO)`,
  
  inputSchema: { /* existing schema */ },
  
  annotations: {
    title: "HiGHS Linear and Mixed-Integer Optimization Solver",
    readOnlyHint: true,
    idempotentHint: true
  }
}

Implementation Details

1. The enhanced description embeds trigger phrases and examples that LLMs can pattern-match
2. Uses MCP-standard fields only (no custom extensions)
3. Structured format within description makes information parseable
4. Adds standard annotations for better tool understanding

Benefits

• Better Discovery: LLMs will recognize optimization problems more reliably
• Clearer Capabilities: Users understand what problems can be solved
• Example-Driven: Concrete examples help map user problems to the tool
• Standards Compliant: Works with any MCP-compliant client

Testing

Create test queries to verify improved discovery:

const testQueries = [
  "I need to find the best way to allocate my budget",
  "How can I minimize shipping costs?",
  "What's the optimal production schedule?",
  "Find the most efficient route for deliveries"
];

Measure whether LLMs correctly suggest the HiGHS tool for these queries.

Related

• See ENHANCEMENT_PLAN.md for broader enhancement strategy
• Builds on Enhance MCP server manifest with metadata for better LLM discovery #45 (manifest enhancement discussion)