Let your AI agent manage your database connections
Connect the coding agent you already use to FluentDB. It can list, create, edit, and delete your database connections for you. You approve every change, and your passwords never reach the model.
An agent creating, listing, and deleting connections over MCP. Every change is approved in FluentDB, and no password ever reaches the model.
What FluentDB's MCP server is
FluentDB is a native macOS database client for PostgreSQL, MySQL, SQL Server, and SQLite, with AI built in as a first-class surface. It ships an MCP server, the standard way for an AI coding agent to call tools inside an app.
You connect your own agent once, and it can drive FluentDB on your behalf. The server runs locally only, bound to 127.0.0.1, is off by default, and is protected by a bearer token. Nothing is exposed to the network or the cloud.
Your agent could already use your connections, opening them and proposing SQL for you to run. The new capability is managing the connections themselves: the plumbing, not just the queries.
Four tools, spoken in plain language
You talk to your agent the way you always do. Behind the scenes it calls these tools in FluentDB, and you stay in control of what lands.
See every connection
“List all connections available in FluentDB”
Returns every connection with full, non-secret detail: name, group, engine, host, port, database, username, SSL mode, read-only flag, SSH summary, timestamps, and whether a password is on file. The passwords and keys themselves are never returned.
Set them up from a file or a description
“Create all connections based on this file connections.json”
FluentDB builds each connection, files any passwords securely in your Keychain, and they appear in your sidebar, ready to open. No form to click through.
Change only what you name
“Update all connections”
Only the fields you name change. Everything else is left exactly as it was, so an edit is never a surprise rewrite.
Remove them, cleanly
“Remove all postgres connections”
Removes the connections and cleans up after them. Stored passwords, SSH secrets, and any saved queries all go too.
Built for batches
Every write is batch-first. Create ten connections, update five, delete three, all in one request. Your agent describes the whole batch and you approve it once. If any item is malformed, FluentDB rejects the entire batch up front with a precise, per-item reason, so you are never asked to approve something that would fail.
Every engine, every auth style
The setup capability is specified in full, so your agent knows exactly what is possible and can stand up even complex, real-world connections in a sentence.
Engines
- PostgreSQL
- MySQL
- Microsoft SQL Server
- SQLite, including encrypted SQLCipher files
Connectivity and authentication
- Direct TCP with username and password.
- TLS modes, from off, to opportunistic, to full certificate and hostname verification.
- SSH tunnels and bastions with three auth styles: a password to the bastion, a private key (ed25519 or RSA) with an optional passphrase, or OpenSSH config.
- OpenSSH config delegates to your system ssh and honors ~/.ssh/config, so jump hosts (ProxyJump) and AWS SSM Session Manager bastions just work.
- Read-only connections, set at creation, so your agent can spin up a connection built for safe querying.
- Groups and folders, so agent-created connections land tidy, not in a junk drawer.
How firmly read-only is enforced depends on the engine. It is a hard guarantee on PostgreSQL and SQLite, a session-level flag on MySQL, and advisory on SQL Server, which cannot enforce it at the connection level.
Security first, because this touches credentials
Secrets never reach the AI
Passwords, SSH secrets, and SQLite passphrases are passed in and written straight to the macOS Keychain. They are never stored in plain data, never returned by any tool, and never shown in any prompt. Your agent sees only that a password will be stored, never the value.
Nothing changes without your approval
Create, update, and delete are each blocked until you click Approve. FluentDB comes to the foreground to ask, even if it was in the background, and shows you exactly what will change. For an edit, that is a field-by-field before and after diff.
Local and token-authenticated
The server listens on loopback only and requires a bearer token. It is off until you turn it on, and you can revoke every setup at once by regenerating the token.
The agent proposes, you dispose
Your agent can do the tedious assembly, typing out hostnames, ports, SSL modes, and SSH bastion details from a plain description. What actually lands stays under your control.
It is the same idea as FluentDB's existing flow, where the AI writes the SQL and you click Run, now extended to the connections themselves. Convenience without handing an AI unsupervised access to your production databases.
What people use it for
Recreate my connections on a new laptop
Describe your setup, or point your agent at a config, and rebuild the whole set in one approved batch.
Set all my staging connections to read-only
One batch update flips the flag across every matching connection, with a diff you approve before it applies.
Add a Postgres connection behind our SSM bastion
Described in a sentence. OpenSSH-config auth means AWS SSM and ProxyJump jump hosts work without a form to fill in.
Works with the agents you already use
Any MCP client that speaks HTTP: Claude Code, OpenAI Codex, Cursor, Windsurf, Gemini CLI, and more. FluentDB generates the exact setup snippet for your client under Settings, in the Coding Agents (MCP) tab. Connect once and you are set.
Frequently asked questions
Can the AI see my passwords?
No. Secrets go straight to the macOS Keychain and are never returned to the agent or shown in any prompt.
Can the AI change my connections without asking?
No. Every create, edit, and delete requires your explicit Approve. FluentDB shows you exactly what will change first, including a field-by-field diff for edits.
Which databases are supported?
PostgreSQL, MySQL, Microsoft SQL Server, and SQLite, including encrypted SQLite files.
Does it work with SSH bastions?
Yes. Password, private key, and OpenSSH-config modes are all supported, and OpenSSH config covers jump hosts (ProxyJump) and AWS SSM Session Manager.
Which AI tools can connect?
Any MCP client over HTTP, including Claude Code, OpenAI Codex, Cursor, Windsurf, and Gemini CLI. FluentDB generates the setup snippet for each in Settings.
Is it cloud-based?
No. The server runs locally on your Mac, bound to loopback only, off by default, and protected by a token.
Stop wrestling
with your database.
Download FluentDB and run your first AI-assisted query in 60 seconds.