kc-mcp

Offline Knowledge Center tools for coding agents — search, cite, and quiz over lesson notes (MCP-shaped).

Who it's for: Azure AI SAs, platform engineers, and educators who want agents to cite their teaching content.

Why this exists

Solution Architects teach constantly (reels, labs, ADRs). Agents forget that context unless it is a tool. kc-mcp turns a local teaching corpus into MCP-style tools so Cursor / Claude / custom agents can search with citations, pull clip notes, list topics, and generate offline quizzes — no vector DB bill, no cloud round-trip for the happy path.

Related MCP server: Study Prep MCP Server

Install

python -m venv .venv && source .venv/bin/activate
pip install -e .
# optional MCP SDK transport:
pip install -e ".[mcp]"

Or with pipx (once published to PyPI): pipx install kc-mcp — until then use editable install from this repo.

30-second demo

python -m kc_mcp --help
python -m kc_mcp topics
python -m kc_mcp search "key vault managed identity"
python -m kc_mcp quiz "MCP"

Or simply:

make demo

What it is NOT

• Not a production vector database or hosted RAG platform

• Not connected to Instagram, YouTube, or Azure by default

• Not embedding-based (MVP uses keyword / TF-IDF over Markdown)

Architecture

Roadmap

• Optional embedding backend behind the same tool schemas

• Ingest real reel transcripts into sample_corpus/

• Richer MCP resources (per-lesson URI)

Contributing

See CONTRIBUTING.md. Be kind — CODE_OF_CONDUCT.md. Security reports: SECURITY.md.

MCP / Cursor plug-in

Install the optional MCP extra, then point Cursor (or any MCP host) at the stdio server:

pip install -e ".[mcp]"

Example ~/.cursor/mcp.json entry:

{
  "mcpServers": {
    "kc-mcp": {
      "command": "python",
      "args": ["-m", "kc_mcp.mcp_server"],
      "cwd": "/absolute/path/to/kc-mcp"
    }
  }
}

Without the SDK, list tool schemas via CLI (tools subcommand where available) or see src/kc_mcp/ for the JSON-RPC-shaped tool table.

License

MIT