Use MCPs when they reduce hallucinations or automate repo-external lookups. This project is Python CLI + PDF + LLM routing; not all generic MCPs add value.
High value for this stack
MCP / pattern
When to use
Context7 (resolve-library-id → query-docs)
Current API for LiteLLM, Pydantic v2, PyMuPDF, pytest, Ruff, MkDocs Material, httpx, etc. Prefer over stale training cutoffs.
Browser / IDE browser
After MkDocs or marketing doc edits: smoke navigation, search, and anchors.
Optional (team-dependent)
MCP
When to use
Atlassian (Jira/Confluence)
If your org tracks specs, bugs, or runbooks there; link issues in PRs, not secrets.
Low value / out of scope for core DocTranslater work
Cloudflare Workers/Pages/D1/R2 MCPs — unrelated unless you add a Cloudflare deployment to this repo.
Laravel Forge — unrelated to this codebase.
Hygiene
Do not paste API keys or proprietary PDFs into MCP queries; use env vars and redacted examples.
Prefer repo files as source of truth for behavior; use Context7 for upstream library semantics.