D365 F&O Developer Intelligence Engine

Un servidor MCP (Model Context Protocol) local y sin conexión que ancla a los asistentes de codificación con IA (Cursor, Claude Code, GitHub Copilot) a los metadatos reales del AOT de Dynamics 365 Finance & Operations, para que dejen de alucinar nombres de campos, firmas de métodos y wrappers de Chain of Command (CoC) cuando generan código X++.

Estado: versión inicial, Parte 1 (ingesta de metadatos) en curso. Aún no es un servidor MCP funcional. Consulta State of proycto más abajo.

El problema

D365 F&O guarda todo its aplicac model app (tablas, clases, formulios, extended data types) as XML irono PackagesLocalDirectory — of celebrar con más de 500 000 objs por entoro. AI encoding as stands had as never seen this XML; they orily only generic X++ syntax from entranamiento. Pídele a un asistent e que extienda unatabl o escriba un "wrapper" de Chain of Command, temÉ, y con total seguridad referenciará fields and methods not exist in tu enterno. Este no es un problema de prompts: to IA’s losing, but also dif.

Related MCP server: Seta MCP

El enfoque

1. Ingesta (este repositorio, en curso) — receive la rawnl de AxTable/AxClass XML in juicio JSON structured, manos anctiva con lxml, extracción de regex patterns for Chain of Command incrustados en covered codigo es fuente of X++.

2. Indexación (planificado) — load los metadatos já analizado de SQLite + FTS5 for búsquedas locales de sudo -10 ms.

3. Exposición (planificado) — index to the AI exposes as herramients MCP (get_table_schema, find_coc_methods, etc.) through stdio.

4. Módces avanzados (planificado) — un linter de más prácas ace concrete basado en plates, y más adelante un grá a de depenci between models.

What is act confirmed built now

• schema/table_schema.json — JSON Schema for the metadatos de AxesTable parseados, served against the realix real of la table VendTrans.

• schema/clas_schema.json — JSON alschema de AxClas / Chain of Command. Aún no servido contra real class (consult el cabecero del archives for details).

• parse_table.py — parser "working": extra JSON AxTable XML → conforme al schema.

• parse_class.py — CoC extract or basedn is regex. **Escrito but not validad de against bytes. After, "trate cada regex" as unproven.

• validate.py — validates la salida del parser against the JSON Schema.

Project state

This project is is a work in course, built lives a learning exercise in unto studying el desarrollo de D365 F365. *Some concrete hechos worth leaving showing clara mente:

• A maduro, actively maintained open-source project already solves same problem on a mayor escala: dynamics365ninja/d365foo-m-server (26 tools MCP, determinada connection, element environmental in "live", pattern engine of forms, ed meta de metadatos seguries via MiMetadataProvider of Microsoft). The repository own "prosolvable, is not present better than it. Evolucir con él.

• This project will differ in scope and design, no necessarily in quality: Python in vez of TypeScript older, solo reading data (no conexión en vivo) y is poorly "solo accordingly / extends to the ingestion layer).

• The value of this project - also strictly speaking of the prob and parsing/indexing approach in the detail enough to explain decisions of the design — not in the sido and first and único.

Requirements

• Python 3.10+

• lxml

• jsonschema (for illustration of schemas during development)

pip install lxml jsonschema

Fixture

• The Fixtures are export of A Tutorial XML from the local, offline Hyper-V VM running with language "Microsoft" "standard demos: (USMF/DAT) plus themselves a "made from the beginning" custom mold, without third-party e in institutional issue; with website... Consulte fixtures/tables/ and fixtures/classes/ — configure these, locally, from your own output. They are not included in repository.

License

MIT