SQL Buddy

A read-only MCP server that lets your AI agent (Claude Code, and any MCP-compatible client) safely query your MySQL and ClickHouse databases.

Your agent asks in plain language; SQL Buddy runs the SQL, enforces read-only (no INSERT/UPDATE/DROP ever reaches your data), adds safety LIMITs, and caches results. You stay in control — the agent can look, never touch.

Your AI agent  ──MCP──►  SQL Buddy (runs locally)  ──►  your databases
                          • opens its own SSH tunnel through a bastion
                          • read-only guard · auto LIMIT · schema lookup

SQL Buddy runs locally (a small Python program, not a cloud service). It opens the SSH tunnel itself from a bastion + key + port mapping in db.ini — no separate ssh -N terminal, and your credentials never leave your machine.

Prerequisites

• Python 3.11+ — python3 --version
• uv (recommended) or plain pip.
• SSH access to a bastion/jump-host that can reach your databases: its user@host + ssh port (often a high port like 22222, not 22), a private key file, and each DB's host:port as seen from the bastion.
• Read-only DB credentials — a user that can SELECT but not write.
• An MCP-compatible client (e.g. Claude Code, Trae).
• Windows only (Auto tunnels): OpenSSH Client. (Skip if you run the tunnel yourself in another terminal.)

Deep notes on tunnels, Windows/Trae, and every db.ini field live in docs/setup.md and docs/windows-trae.md.

Quickstart

Install (current release 0.3.3, Python ≥ 3.11):

# From PyPI (recommended):
pip install sqlbuddy-nwi          # provides the `sql-buddy` command
# …or run the latest instantly without installing:
uvx --from sqlbuddy-nwi sql-buddy

# From source (for development):
git clone https://github.com/yyx462/sql-buddy && cd sql-buddy
uv run server.py                  # auto-provisions an isolated .venv, then serves

Configure & connect:

sql-buddy init                    # interactive wizard → writes connections/config/db.ini
sql-buddy doctor --connect        # green-checks db.ini, tunnels, MCP client config
sql-buddy mcp add                 # registers sql-buddy with Claude Code (writes .mcp.json)

From a git checkout, prefix these with uv run (e.g. uv run sql-buddy init). From a pip / uvx install, run sql-buddy directly — it's on your PATH.

Restart your client, then ask your agent "list the tables in the ck connection" — if it answers, you're live. 🎉

Bare sql-buddy (no command) runs the MCP server — that's what your AI client invokes, so configs hold no fragile absolute .venv/bin/python path. sql-buddy mcp add auto-detects how you installed and writes the right command: uvx --from sqlbuddy-nwi sql-buddy for a pip/uvx install, uv --directory … run server.py for a checkout. Other commands: version, doctor [--connect], mcp print|add, update-check, init. On Windows, see docs/windows-trae.md and use sql-buddy mcp add --client trae.

Configure connections (db.ini)

Run the wizard (easiest) or copy the template:

sql-buddy init
# or, from a git checkout, copy the template and edit by hand:
cp connections/config/db.ini.example connections/config/db.ini
$EDITOR connections/config/db.ini

Each database is one section; the [bracketed] name becomes the connection name your agent uses (e.g. ck, mysql, prod-read).

# db.ini  —  NEVER commit this file (gitignored). It holds live credentials.

[ssh:prod]
host = bastion.example.com
port = 22
user = your_ssh_user
key  = ~/.ssh/id_ed25519          # path to your private key (~ expanded)

[ck]
type        = clickhouse
tunnel      = ssh:prod            # self-managed tunnel through the bastion above
remote_host = 10.0.0.21           # DB host AS SEEN FROM the bastion
remote_port = 9000                # DB port AS SEEN FROM the bastion
local_port  = 10021               # localhost port the driver connects to
user        = your_readonly_user
password    = your_password
database    = default

[local-db]                        # direct (non-tunneled) connection also works
type     = mysql
host     = localhost
port     = 3306
user     = your_readonly_user
password = your_password
database = myapp

remote_host/remote_port are the DB as seen from the bastion, not your laptop — getting this wrong is the #1 setup failure. For the full field reference, SSH troubleshooting, and External (ssh -N) mode, see docs/setup.md.

🔒 Double-check git status does not list db.ini before any commit.

Connect your AI agent

sql-buddy mcp add                      # Claude Code project → ./.mcp.json
sql-buddy mcp add --scope user         # Claude Code user    → ~/.claude.json
sql-buddy mcp add --client trae        # Trae project        → ./.trae/mcp.json
sql-buddy mcp add --client trae --scope user   # Trae user-global

Restart your client. Preview the exact block first with sql-buddy mcp print (paths filled in):

{
  "mcpServers": {
    "sql-buddy": {
      "command": "uv",
      "args": ["--directory", "/absolute/path/to/sql-buddy", "run", "server.py"]
    }
  }
}

No uv on the machine that runs your MCP client? sql-buddy mcp print --python sets command to the running interpreter (sys.executable). For other MCP clients (Cursor, Continue, WorkBuddy, …) and Windows/Trae specifics, see docs/setup.md and docs/windows-trae.md.

Security model (short)

• Read-only enforced in code, before any SQL reaches the network. Write keywords (INSERT, UPDATE, DELETE, DROP, ALTER, TRUNCATE, …) are rejected by a guard layer wrapping every connection.
• Credentials stay local — in db.ini on your machine, never sent anywhere.
• SSH tunnels bind to localhost only (127.0.0.1); key-based auth only.
• LIMIT is always applied to SELECT/WITH (default 10, hard cap 2000).

Full security posture and an error/troubleshooting matrix: docs/reference.md.

Documentation