SQLite MCP Server

What this server does

Point this at a `.db` file and the model can run SQL against it, read and write. Six tools: `read_query` (SELECT), `write_query` (INSERT/UPDATE/DELETE), `create_table` (DDL), `list_tables`, `describe-table`, and `append_insight`. The last one is unusual: it appends a string to a `memo://insights` resource that the host can re-read, so the model can build up a running narrative of what it has discovered as it works through the data. A built-in prompt called `mcp-demo` takes a business domain as its `topic` argument, generates a schema, populates sample data, and walks you through an analysis. It is a tutorial fixture, not a production workflow, but it is a fast way to see what the BI loop is supposed to feel like. Unlike the Postgres MCP server, which is read-only by construction, this one is read-write. `write_query` and `create_table` modify the database directly. There is no transaction isolation across tool calls. Each query commits when it returns. If you only want analytics, do not give the model the path to your production SQLite file. Best fit: an analytics scratch DB, a SQLite dump pulled from production for one-off questions, or learning what the model can do with the included BI prompt template. The reference is archived but the protocol surface is stable; queries that work today will work tomorrow.

Tools included

• read_query

  Run a read-only SELECT query.

• write_query

  Run an INSERT / UPDATE / DELETE statement.

• list_tables

  List all tables in the database.

• describe_table

  Show the schema of a table.

The authoritative list comes from the server's tools/list at runtime. This catalog entry may lag the upstream until the next ToolPiper release.

Install in ToolPiper

ToolPiper handles the JSON configuration. Click Install, fill in any required tokens, and the server is ready in your chat.

Install in another client

{
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": [
        "mcp-server-sqlite",
        "--db-path",
        "${SQLITE_DB_PATH}"
      ]
    }
  }
}

Example prompts

Security notes

This server is read-write by default. Anything the model can express in SQL it can execute, including `DROP TABLE`. Point it at a copy of your data, not the original, unless you trust the model's discipline. There is no SQL parser between the model and SQLite; the query goes straight through. The database file lives wherever `--db-path` points, so secrets in that file are visible to the assistant. For production analytics, prefer Postgres with a read-only role and the Postgres MCP server. The upstream reference is archived; no official replacement has shipped, though community forks exist.

Related tools