| name | superlazy-mcp |
| description | Build MCP servers with minimal tokens. Use when creating or modifying an MCP (Model Context Protocol) server integrating any external API or service, in TypeScript or Python. |
Superlazy MCP Builder
SUPERLAZY mode: output ≤3 lines (questions / user-actions / ✅❌); minimal code, full quality; do the work, never narrate it.
Defaults (proceed silently)
- TypeScript + official
@modelcontextprotocol/sdk (Python + FastMCP only if the project is already Python).
- Transport: stdio for local; streamable HTTP with stateless JSON for remote.
- Docs beat memory: fetch
https://modelcontextprotocol.io/sitemap.xml, then needed pages with .md suffix; SDK README from its GitHub raw. Do this before writing code, silently.
Tool design rules
- Comprehensive API coverage first; workflow-convenience tools only when the API composition is genuinely painful.
- Names:
service_verb_noun (github_create_issue). Action-oriented, consistent prefix.
- Schemas: Zod (TS) / Pydantic (Py), constraints + descriptions + examples on every field;
outputSchema/structuredContent where supported.
- Every listing tool paginates and filters — tools return focused data, not dumps. Small default page size (~20); concise text content, big payloads in
structuredContent only, never duplicated as prose.
- Errors are actionable: what failed + what the agent should do next.
- Annotations on every tool:
readOnlyHint, destructiveHint, idempotentHint, openWorldHint.
- Shared infra once: API client + auth, error helper, response formatter. DRY.
Verify before ✅
TS: npm run build exit 0. Py: python -m py_compile. Then exercise the server (MCP Inspector or direct calls) — at least one real call per tool category. Evidence per superlazy-verify.
