JCGE Agent Interface MCP server

JCGEAgentInterface

<!-- mcp-name: io.github.equicirco/JCGEAgentInterface.jl -->

What is a CGE?

A Computable General Equilibrium (CGE) model is a quantitative economic model that represents an economy as interconnected markets for goods and services, factors of production, institutions, and the rest of the world. It is calibrated with data (typically a Social Accounting Matrix) and solved numerically as a system of nonlinear equations until equilibrium conditions (zero-profit, market-clearing, and income-balance) hold within tolerance.

What is JCGE?

JCGE is a block-based CGE modeling and execution framework in Julia. It defines a shared RunSpec structure and reusable blocks so models can be assembled, validated, solved, and compared consistently across packages.

What is this package?

MCP-compatible interface layer for agents that need to discover, load, solve, validate, and inspect

JCGE models. It exposes a small action API and a standard MCP stdio server over the same functionality.

The agent surface is intentionally focused on:

• discovering installed JCGE packages and available functionality,
• listing and describing reusable JCGEBlocks components,
• guiding users through CGE model development with JCGE blocks,
• guiding formulation choices such as equality systems, inequalities, MCP/complementarity, and optimization-style representations,
• guiding source-data preparation with released JCGEImportData adapters while retaining model-owned SAM construction choices,
• guiding calibration with the currently available JCGECalibrate loaders, SAM helpers, labeled containers, and parameter workflows,
• guiding solver choice and reporting from generated equations,
• listing released JCGEExamples reference model modules,
• solving registered RunSpec models,
• validating solved contexts,
• rendering implemented equations, blocks, and symbols through JCGEOutput,
• reporting or applying updates to released JCGE packages in the active Julia environment.

The package services are grouped in the documentation as discovery, modeling guidance, model interaction, reporting, environment maintenance, and MCP runtime services. See docs/src/services.md for the full service contract and current limits.

MCP server

The standard MCP transport is:

using JCGEAgentInterface
serve(transport=:mcp_stdio)

Minimal local MCP client configuration:

{
  "mcpServers": {
    "jcge": {
      "command": "julia",
      "args": [
        "--project=/path/to/JCGEAgentInterface.jl",
        "-e",
        "using JCGEAgentInterface; serve(transport=:mcp_stdio)"
      ]
    }
  }
}

The MCP tools use the jcge_ prefix:

• jcge_capabilities
• jcgelistblocks
• jcgedescribeblock
• jcgemodelingguide
• jcgeformulationguide
• jcgesolverguide
• jcgeimportdata_guide
• jcgecalibrationguide
• jcgereportingguide
• jcgelistexamples
• jcgepackagestatus
• jcgeupdatepackages

The original custom JSON action protocol remains available:

serve() reads JSON requests from stdin and writes JSON responses to stdout.

Docker and MCP Registry

OCI-oriented MCP registry metadata is declared in server.json.

Future version tags matching v..* trigger .github/workflows/publish-mcp.yml, which:

Shortened. The full README is on GitHub.