Trace MCP

# Trace MCP - Core Truth > "Immutable scope boundaries. Reference before adding features." ## 1. Purpose Trace MCP analyzes codebases to detect mismatches between data producers (MCP Tools) and consumers (Frontend/Client Code). ## 2. What It Does (The Yes List) - **Extract Schemas**: From MCP tools, TypeScript types, API responses. - **Trace Flow**: Follow properties through transformations (renames, destructuring). - **Detect Mismatches**: Where producer output `!=` consumer expectation. - **Report Only**: Structured data (JSON) or human summaries (Markdown). ## 3. What It Does NOT Do (The No List) - ❌ **State**: No database, no memory between runs. - ❌ **Fixing**: No codemods, no auto-repairs. - ❌ **Decisions**: No logic to "decide" what to do with findings. - ❌ **Integration**: No direct Slack/GitHub/CI API calls (that is the caller's job). ## 4. Design Principles 1. **Pure Analysis**: `Codebase -> [Trace MCP] -> Structured Findings` 2. **Stateless**: Every run is fresh. 3. **Composable**: Can run just extraction, or just comparison. 4. **Minimal Dependencies**: Do not bind to specific frameworks unless strictly for parsing. ## 5. Success Criteria An agent can call Trace MCP tools and receive actionable data about schema mismatches without Trace MCP needing to know _what_ the agent will do with that information.