Databricks MCP Server

# Product Requirements Document ## Overview **MCP Guessing Game Server** - A minimal viable MCP server that exposes a single prompt-based slash command for a number guessing game. ### Problem Statement We need to validate the FastAPI MCP integration pattern with the simplest possible implementation before building complex Databricks-integrated prompts. ### Solution Create an ultra-simple MCP server that: - Exposes one prompt file (`prompts/guess_number.md`) as a slash command - Implements one FastAPI endpoint to handle the game logic - Validates the end-to-end MCP architecture ## Target Users **Primary User:** Developers testing MCP integration **Use Case:** Validate that FastAPI can expose prompts as Claude Code slash commands ## Core Features ### Single Feature: Number Guessing Game - **Slash Command:** `/mcp__mcp-commands__guess_number` - **Prompt File:** `prompts/guess_number.md` (static markdown file) - **API Endpoint:** Single FastAPI endpoint that accepts a guess and returns response - **Game Logic:** - Server picks random number 1-100 - User guesses via slash command - Server responds "higher", "lower", or "correct" - No state persistence (new game each request) ### Technical Scope - **1 prompt file** in `prompts/` directory - **1 FastAPI endpoint** for game logic - **0 external dependencies** beyond fastapi_mcp - **0 Databricks integration** (future enhancement) ## Success Metrics ### MVP Success Criteria - [ ] Prompt file is discovered and exposed as slash command - [ ] Slash command executes and calls FastAPI endpoint - [ ] Game logic works (higher/lower responses) - [ ] MCP server connects to Claude Code successfully ### Implementation Success - [ ] Code is deployed to Databricks Apps - [ ] MCP server is accessible via network - [ ] Claude Code can connect to MCP server - [ ] User can play guessing game via `/mcp__mcp-commands__guess_number` command ## Implementation Priority **Phase 1: Ultra-Simple MVP** (This Release) 1. Create `prompts/guess_number.md` 2. Add fastapi_mcp integration to existing FastAPI app 3. Implement single guessing game endpoint 4. Test MCP server connection locally 5. Deploy to Databricks Apps and test slash command **Phase 2: Future Enhancements** (Later) - Add Databricks SQL query prompts - Add MLflow model interaction prompts - Add workspace file management prompts - Support prompt arguments and parameters ## Technical Requirements ### Minimal Architecture - **Framework:** Extend existing FastAPI app with `fastapi_mcp` - **Prompts:** Single markdown file in `prompts/` directory - **State:** Stateless (no persistence required) - **Dependencies:** Add `fastapi_mcp` to existing `pyproject.toml` ### Acceptance Criteria 1. User types `/mcp__mcp-commands__guess_number` in Claude Code 2. Prompt is executed and calls our FastAPI endpoint 3. Server responds with game feedback 4. Integration validates MCP architecture works end-to-end