MCP Mifosx

What openMF/mcp-mifosx MCP server does

openMF/mcp-mifosx MCP server exposes banking operations from Apache Fineract and the Mifos X ecosystem through the Model Context Protocol. It is intended to sit between an MCP client and a Fineract REST API, allowing an AI assistant or workflow agent to request typed operations instead of constructing backend HTTP calls directly.

The repository covers common core-banking domains, including clients, identities, groups, centers, loans, savings, collateral, staff, accounting, charges, documents, reports, and financial statistics. Some implementations also include composite or bulk operations. The exact surface depends on the language implementation selected: Go lists 102 tools, Rust 89, Python 49, and Java 38.

How it works

The server is designed as a stateless integration layer. It does not retain user data, personally identifiable information, or AI memory. Each implementation targets the same Apache Fineract REST API and aims to keep field routing and returned JSON structures aligned across languages.

Four implementations are included:

• Go provides native tooling with stdio and optional SSE transport.
• Rust uses asynchronous I/O and adds bulk-operation support.
• Python uses FastMCP with a domain-oriented layout.
• Java uses Quarkus and separates backoffice operations from user recommendations.

The MCP client remains outside this repository. The README describes use with MCP-compatible systems, including Claude Code, claude.ai, n8n, and custom workflow agents. The server can therefore be paired with a client or agent framework that supports the chosen MCP transport.

Setup and configuration

Each implementation has its own build and runtime requirements. Go requires Go 1.21 or later; Rust requires Rust and Cargo; Java requires JDK 21 or later plus Maven; and Python requires Python 3.10 or later and pip.

The Go and Rust setup instructions use implementation-specific .env.example files. Go can run over stdio by default and switch to SSE when PORT is set, while Rust runs its compiled release binary. Java accepts a Fineract base URL, a basic-auth token, and a tenant identifier through MIFOSX_BASE_URL, MIFOSX_BASIC_AUTH_TOKEN, and MIFOS_TENANT_ID. Python installs dependencies from its requirements file, reads configuration from .env, and starts with python mcp_server.py.

Tools and capabilities

The Go implementation has the broadest listed coverage, including 16 client and identity tools, 26 document and report tools, 23 loan and savings tools, 13 group and center tools, 19 bulk or composite tools, and 5 accounting or statistics tools.

Rust emphasizes clients, loans, groups, savings, centers, staff, accounting, charges, and bulk processing. Python groups its 49 tools around clients and groups, loans and savings, and staff and accounting. Java provides 21 backoffice tools and 17 recommendation-oriented tools.

Examples in the repository include creating and activating clients, creating loan products, submitting and approving loan applications, and disbursing loans. Availability of a specific operation depends on the implementation in use.

Limitations and notes

Tool counts and capabilities are not identical across the four implementations. The repository describes functional parity as an aim rather than a guarantee that every tool exists everywhere. You must also supply access to an Apache Fineract or Mifos X instance and configure the selected implementation's credentials and endpoint.

The project is an integration tier, not an AI client, user interface, memory store, or standalone banking backend. The MCP-compatible client and its orchestration remain separate from the repository. MCP Inspector examples are provided for testing, but Inspector is a debugging utility rather than the server itself.