Memory-Alpha-MCP CLAUDE.md

Memory Alpha MCP Server

Project Context

MCP server providing Star Trek knowledge from Memory Alpha's wiki API. TypeScript, ESM, Node 18+.

Primary libraries and runtime conventions:

• @modelcontextprotocol/sdk for MCP server/tool/prompt/resource registration
• wtf_wikipedia for primary wikitext parsing
• zod for tool schemas and input validation
• vitest for unit tests
• NodeNext ESM import style with explicit .js extensions on relative imports

The server runs over stdio and fetches Memory Alpha content dynamically at runtime. Treat transport safety and attribution as core requirements, not optional polish.

Commands

Run from the repository root:

• npm install
• npm run build
• npm run dev
• npm test
• npx @modelcontextprotocol/inspector node dist/index.js

Non-Negotiable Rules

• Never use console.log because it can corrupt the stdio MCP transport. Use console.error via src/utils/logger.ts.
• All successful tool responses that include Memory Alpha content must use withAttribution() from src/utils/attribution.ts.
• Error responses that contain no wiki content should not add attribution.
• All string tool parameters need .max(...) bounds in their zod schema.
• Keep content dynamic. Do not embed Memory Alpha article text into the repository.
• Preserve NodeNext ESM imports with .js extensions on relative paths.
• wtf_wikipedia quirk: call section.text({}) and doc.text({}), not .text().

Architecture Notes

• src/index.ts wires up the MCP server, stdio transport, graceful shutdown, and registration of tools/prompts/resources.
• src/api/client.ts owns fetch behavior: rate limiting, timeout, retry, and in-memory caching. Reuse it instead of rolling custom HTTP logic.
• src/parser/wikitext.ts and src/parser/infobox.ts contain the parsing edge cases for disambiguation handling and nested sidebar templates.
• src/resources/index.ts contains original static reference material. Keep those resources original rather than copied wiki content.