Skip to content

Every analytics tool wants a subscription. And your data. I built Twiligent instead. YouTube stats. Instagram stats. One dashboard. Runs locally on your machine.

2.9 KB
Raw Blame

MCP Usage - Decision Guide

Four MCP servers are available. Each has a specific purpose. Using the wrong one wastes time and produces worse results than using the right one.

Quick Decision Tree

Need API docs, library reference, or framework docs?
    -> Context7

Need to find a file, symbol, or understand the codebase?
    -> Filesystem (or direct Grep/Glob tools first)

Need to plan a complex feature, debug a hard problem, or think through architecture?
    -> Sequential Thinking

Need to research UX patterns, browser compat, competing products, or best practices?
    -> Parallel Search

When Each MCP is the Right Choice

MCP Use it for Do not use it for
Context7 Hono API, Supabase JS methods, Instagram Graph API params, Cloudflare Workers API, Recharts props, shadcn component props Writing code, debugging business logic, code review, general programming
Filesystem Finding where a function is defined, locating all files that import a module, moving/renaming across multiple files Writing code, understanding architecture (read the docs instead), large-scale analysis
Sequential Thinking Planning a multi-step feature before implementing, architecture decisions, debugging when root cause is not obvious Lookups (use Context7), simple one-step tasks
Parallel Search UX research for a new feature, browser API compatibility, finding how competitors handle a UI problem, finding accessibility guidelines Library docs (use Context7), codebase exploration (use Filesystem)

Workflow Examples

Adding Instagram Reels support to the publisher

  1. Sequential Thinking - plan the full implementation (backend params, frontend UI changes, DB schema changes if any)
  2. Context7 - look up Instagram Graph API Reels container params (media_type=REELS, required fields)
  3. Parallel Search - research UX patterns for Reels upload interfaces
  4. Implement

Debugging a Supabase query that returns wrong data

  1. Filesystem - find all callers of the function to understand the context
  2. Context7 - look up Supabase select(), eq(), single() semantics if the issue might be a library API misunderstanding
  3. Sequential Thinking - if the root cause is still unclear, walk through the data flow step by step

Adding a new shadcn component

  1. Context7 - get the exact props and composition for the component
  2. Direct implementation - no other MCP needed

Deciding whether to add a new npm dependency

  1. Sequential Thinking - analyze the trade-offs (bundle size, alternatives, maintenance)
  2. Parallel Search - check if there is a lightweight alternative or a Web API equivalent

Detailed Docs

  • docs/mcp/CONTEXT7.md - when, how, and example queries
  • docs/mcp/FILESYSTEM.md - file search, symbol location, refactoring
  • docs/mcp/SEQUENTIAL_THINKING.md - planning and debugging
  • docs/mcp/PARALLEL_SEARCH.md - research workflows