# SAGE — Solver-Augmented Grounding Engine [Health: Active]

**Category:** 🧮 Data Science Tools  
**Repository:** https://github.com/pragnakar/Project_Sage  
**GitHub Stars:** 0  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/sage-solver-augmented-grounding-engine

## Description
LLM-native optimization solver: LP, MIP, portfolio & scheduling via HiGHS/OR-Tools.

## Tools
Capabilities this server exposes over MCP:

- **solve_optimization** — Solve LP / MIP / portfolio / scheduling from JSON
- **read_data_file** — Read an Excel or CSV file and return a preview
- **solve_from_file** — Read + solve + write results in one step
- **explain_solution** — Narrate the most recent solve result
- **check_feasibility** — Check feasibility; if infeasible, compute IIS
- **generate_template** — Create a blank Excel template for a problem type
- **suggest_relaxations** — Rank constraint relaxations for the last infeasible result

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "sage-solver-augmented-grounding-engine": {
    "command": "uvx",
    "args": ["sage-solver-mcp"]
  }
}
```

## Documentation

## What SAGE — Solver-Augmented Grounding Engine does

SAGE — Solver-Augmented Grounding Engine exposes mathematical optimization through MCP for Claude Desktop and other MCP-compatible agents. It accepts structured JSON problem definitions or reads supported spreadsheet files, then returns solver results that can be inspected and explained.

Supported problem categories include linear programming, mixed-integer programming, portfolio optimization, and workforce scheduling. The implementation uses HiGHS for LP and MIP problems and OSQP for quadratic programming. File operations support Excel workbooks and CSV files.

The tool set covers the full workflow: creating a blank input template, previewing data, solving a problem, writing results, explaining a recent result, checking feasibility, and ranking possible constraint relaxations after an infeasible solve. Results can include objective values, variable assignments, feasibility status, dual values, reduced costs, and allowable ranges where applicable.

## How it works

For a direct solve, an agent sends a JSON description containing variables, bounds, constraints, and an objective to `solve_optimization`. File-based workflows use `read_data_file` to inspect workbook or CSV contents, or `solve_from_file` to read input, solve the model, and write the output in one operation.

`check_feasibility` tests whether the stated constraints can be satisfied. If the model is infeasible, SAGE can compute an irreducible infeasible subsystem (IIS), helping identify constraints that conflict. `suggest_relaxations` then ranks possible changes for the most recent infeasible result. `explain_solution` turns the latest solve result into a plain-language account, including relevant sensitivity information when available.

This division keeps language interpretation with the MCP client while delegating feasibility and optimization calculations to dedicated solvers. The repository describes the architecture as hybrid: the language model handles ambiguity and the solver handles mathematical validation.

## Setup and configuration

The documented development installation clones the repository and installs the `sage-solver-core` and `sage-solver-mcp` packages in editable mode.

Claude Desktop can launch SAGE through `uvx` with a server entry whose command is `uvx` and whose argument is `sage-solver-mcp`. After adding the entry to the Claude Desktop configuration, restart the application to expose the tools. If `uvx` is not on the application PATH, the configuration can use its full filesystem path. The README identifies separate configuration locations for macOS and Windows.

SAGE — Solver-Augmented Grounding Engine is described as local software and does not list required environment variables. Example prompts cover a direct staffing LP and a portfolio workbook.

## Tools and capabilities

- `solve_optimization`: solve LP, MIP, portfolio, or scheduling models from JSON.
- `read_data_file`: inspect an Excel or CSV file and return a preview.
- `solve_from_file`: read model data, solve it, and write results.
- `explain_solution`: narrate the most recent solution.
- `check_feasibility`: test feasibility and compute an IIS for infeasible models.
- `generate_template`: create a blank Excel template for a supported problem type.
- `suggest_relaxations`: rank constraint changes for the last infeasible result.

## Limitations and notes

The repository labels SAGE — Solver-Augmented Grounding Engine version 0.1.3 as alpha.

The examples show local file paths and workbook-based workflows, so file access depends on the environment where the MCP process runs. Treat those capabilities as architectural claims rather than documented tool interfaces.

_Full upstream README: https://allmcps.com/mcp/sage-solver-augmented-grounding-engine/readme_