xzwyn/mcp_server

Local MCP TDF Updater (Skeleton)

Overview

• Local-only, free MCP server and Python pipeline to:
  • Parse requirements CSV (you supply mapping).
  • Analyze C++ source for explicit tags and lexical relevance.
  • Parse legacy .tdf in a round-trip fashion preserving formatting.
  • Use a local LLM (your API) to propose updates per TESTID.
  • Merge updates into a fully rewritten .tdf.
  • Generate a Markdown report.

Quick start (Windows)

1. Python env:

   • Create venv:
     • py -3.11 -m venv .venv
     • ..venv\Scripts\activate
   • Install requirements:
     • pip install -r python\requirements.txt

2. Node MCP server:

   • cd mcp-server
   • npm install
   • Launch (for manual test): node mcp-tdf-server.mjs
   • Typically your VS Code MCP client launches this file.

3. Orchestrate (manual, without MCP):

   • ..venv\Scripts\python.exe python\tdf_pipeline\cli.py orchestrate-update-all ^ --csv testdata\requirements.csv ^ --src testdata\src ^ --tdf testdata\sample.tdf ^ --out-tdf artifacts\outputs\updated.tdf ^ --report artifacts\reports\update_report.md

Artifacts

• artifacts\cache: cached parsed CSV, code index, parsed TDF
• artifacts\mapping: TESTID → requirement mapping JSON
• artifacts\contexts: per-TESTID context bundles
• artifacts\proposals: per-TESTID LLM proposals
• artifacts\diffs: diffs for updated fields/regions
• artifacts\outputs: updated.tdf and a backup of the original

Notes

• Wire your local LLM endpoint in python\tdf_pipeline\llm_orchestrator.py (OpenAI-compatible stub included).
• TDF parsing anchors on lines beginning: "TESTID:"; comments (# ...) are preserved.
• Field names are not required; this skeleton replaces the editable body within each TESTID block while preserving the header and comments.