How to Set Up a SQLite MCP Server With Claude Code

A customer emailed asking whether their data had actually synced or whether something was silently failing on their device. I knew we had a conflict table. I knew it had timestamps. I did not know the exact query off the top of my head, and I definitely didn't want to go write it from scratch while a customer was waiting for an answer. I wanted to ask in plain English, get the row count, and reply. This is the setup that made that possible.

We'll connect your SQLite database (via Turso or a libSQL endpoint) to Claude Code through QueryBear. About fifteen minutes.

Part of the SQLite MCP Server guide.

What you'll need

• A SQLite database accessible through a Turso or libSQL HTTP endpoint (QueryBear connects via @libsql/client over HTTP, not local .db files)
• A QueryBear account
• Claude Code installed and authenticated

Why route through a gateway

Even with a read-only libSQL token, handing your endpoint directly to an AI agent means no per-table visibility control, no column masking, and no audit trail. QueryBear sits in between: SQL parser rejects anything that isn't a SELECT, you choose exactly which tables are visible, you block individual columns, and every query is logged. See the SQLite MCP Server guide for the full security picture.

Connect QueryBear to your SQLite database

1. Open the QueryBear dashboard and click "New connection."
2. Choose SQLite (Turso/libSQL). Paste your Turso URL, which looks like libsql://your-db-name.turso.io, and your auth token.
3. Under "Allowed tables," pick the tables you want Claude to see. In a sync-heavy app this might be sync_events, devices, conflicts, and a few others.
4. Under "Blocked columns," add anything sensitive: auth tokens, user secrets, anything you'd be unhappy seeing printed in a terminal.
5. Save. QueryBear connects over HTTP and confirms the schema is accessible.

Install the QueryBear MCP server in Claude Code

In your terminal:

claude mcp add --transport http querybear https://mcp.querybear.com/mcp

Claude Code walks you through OAuth. After that, three tools are available in every Claude Code session: list_connections, get_schema, and run_query. All three proxy through QueryBear; none of them talk directly to your Turso endpoint.

Ask your first question

Open a Claude Code session. Ask something in plain English:

I asked: "Which sync conflicts are still unresolved in the last 24 hours?"

Claude Code generated this query and ran it through QueryBear:

SELECT id, device_id, conflict_type, created_at
FROM sync_conflicts
WHERE resolved_at IS NULL
  AND created_at >= datetime('now', '-24 hours')
ORDER BY created_at DESC;

Got back 7 rows. The whole round trip took under a second.

Claude calls get_schema first, then writes the query using SQLite's datetime('now', '-24 hours') syntax, then calls run_query. If the query hits a table not on the allowlist or a blocked column, QueryBear returns a readable rejection before anything reaches Turso.

A Claude Code-specific tip

Some questions come up on a schedule, like "any unresolved conflicts right now?" If you're checking that after every deploy or every customer complaint, save the query once. In QueryBear, after running a query you're happy with, click "Save as tool" and give it a name you'll actually remember, like unresolved-conflicts.

In Claude Code:

Use the unresolved-conflicts tool.

Claude calls it directly. Same query, same column order, every time. No drift in how it phrases the WHERE clause. For on-call situations especially, it's worth setting these up before you need them.

Using a different tool?

• Claude Cowork
• Cursor
• Windsurf
• ChatGPT
• Codex
• Gemini

Querying a different database?

• Postgres MCP Server
• MySQL MCP Server