SQLite

Originally at src/sqlite in the monorepo, now in servers-archived. The PyPI package mcp-server-sqlite still receives the occasional maintenance release and is the smallest path to giving an agent a real SQL surface without spinning up a database server. What it does: - Schema introspection for tables, indexes, views, and triggers - Parameterised SELECT, INSERT, UPDATE, and DELETE execution - A built-in "business insight memo" resource that accumulates agent findings across a session - Transactions — the agent can BEGIN, do work, and COMMIT or ROLLBACK explicitly - Works against a single on-disk file, no server process or port to manage - Plays well with WAL mode for concurrent read + single-writer workloads Example: Mount a local analytics.db in WAL mode and point the server at it. A notebook-style agent can answer "which cohort retained best in Q1" with a multi-CTE SELECT, and then write the result into a memos table the human reads later in their editor. Setup: Pass --db-path pointing at the SQLite file. The parent directory must exist and be writable. No auth — the threat model assumes trusted local use. Transport: stdio only. The file is opened per session, so two clients running against the same DB path will each hold their own SQLite connection. Notes: Best for small local databases and notebook-style analysis. For anything multi-user or write-heavy, use the Postgres entry in this registry instead. Because upstream is archived the soft-reliability score here reflects maintenance status, not code quality — the server itself is simple and has shipped fewer than a dozen bugs since first release.