SQLite MCP is the local-first counterpart to Postgres MCP. It points at a single .sqlite or .db file on disk and exposes query + schema-inspection tools. No server, no auth, no network — perfect for analysis on a CSV-imported dataset, prototyping a schema before promoting it to Postgres, or letting the agent query a local app’s SQLite store.
What it produces: query, list_tables, describe_table. Read by default, write available with the right launch flag. Returns rows as JSON — easy for the model to reason about.
Best for: founders running a quick ad-hoc analysis (“import this CSV into SQLite, agent, tell me what’s correlated”), exploring an unfamiliar app’s local DB (Mac apps often store state in SQLite), or building a prototype before committing to a hosted DB.
Skip if: your data lives in Postgres or Supabase — use Postgres MCP or Supabase MCP directly. Don’t manually export Postgres → SQLite to use this skill; it’s a step backward.
Setup gotchas: the database path is a launch arg, not a tool param. npx -y @modelcontextprotocol/server-sqlite path/to/db.sqlite locks the agent to that file. Switching files = restart the server. Also: the server is in servers-archived — Anthropic moved active community work elsewhere, but the basics still work fine for typical use.
Real-world workflow: I use it for one-off “I just exported this from somewhere, what’s interesting?” analysis. Last week: exported a customer-event log to SQLite, asked the agent to find the top funnel-drop step. Agent wrote a 30-line query I’d never have written by hand and returned a clear answer.
Compatible alternatives: Postgres MCP for prod data, Supabase MCP for full Supabase, Excel if the destination is a spreadsheet not a query.
Excellent for prototyping. Don’t ship SQLite to production unless your scale really fits it.