Dataki for Claude Code
Query your database from Claude Code
One command adds Dataki's hosted MCP server to Claude Code. From then on it can read the schema of the databases you have connected to Dataki and run read-only queries while it works, without a connection string on your machine.
- Claude Code
- Dataki MCP server
- Your database, read-only
- Signs in with
- A Dataki API key.
- Works in
- Claude Code, in any terminal.
- You need
- Claude Code, and a Dataki account with at least one data source.
- Server URL
https://dataki.ai/api/mcp
01Set it up
3 steps, a couple of minutes
Dataki's connect page walks through the same steps with buttons, and creates a key wherever one is needed.
- 01
Create a key
Open the Claude Code steps on Dataki's connect page and create a key. The page fills it into the command for you.
- 02
Run this in a terminal
Terminal claude mcp add --transport http dataki https://dataki.ai/api/mcp --header "Authorization: Bearer YOUR_DATAKI_KEY" - 03
Ask it something in Claude Code
Start with
What data can you see in Dataki?.
02Ask it
What to ask in Claude Code
The assistant finds your sources, reads what the columns mean, writes the SQL and hands it to Dataki, which runs it read-only and returns the rows. The query is part of every answer.
Which tables in the billing database hold subscriptions, and how do they join to customers?
Answered from get_data_context: the tables, columns and types, plus any metric definitions and join paths your team has recorded for that source.
How many users signed up each day since Monday's release, and how does that compare with the week before?
The assistant writes the SQL, Dataki runs it read-only and returns the rows. The query is in the answer, so you can check it.
Are there any orders without a matching customer? Show me the ten most recent.
The kind of check you would otherwise run by opening a SQL client against production, without handing the editor a connection string.
Put a chart of daily signups for the last 90 days on a dashboard called Growth.
save_dashboard keeps it as a live Dataki dashboard with a link, which refreshes on its own and can be shared with people who never open an editor.
03What it can reach
Any source you connect to Dataki
Claude Code sees the sources your Dataki account can open, and one question can read more than one of them.
-
PostgreSQL Connects directly
-
MySQL Connects directly
-
Amazon Redshift Connects directly
-
Google BigQuery Connects directly
-
Supabase Connects directly
-
Google Sheets Connects directly
-
CSV and Excel files Connects directly
-
Neon Connects directly
-
PlanetScale Connects directly
-
Google Analytics Through BigQuery
-
Firebase Through BigQuery
-
Google Search Console Through BigQuery
-
Cloud Firestore Through BigQuery
-
Stripe Through BigQuery
-
MongoDB Through a file or BigQuery
04Worth knowing
What Claude Code gets, and what it never does
- The key is a password
- It acts as you, and anything you can reach in Dataki, it can reach. It carries daily ceilings on query count and bytes scanned, every query it runs is logged with its name, and you can revoke it from Dataki at any time. Do not commit it to a repository.
- Not a connection string in a config file
- A local database MCP server needs the database URL on your machine. With Dataki, Claude Code holds a revocable key; the database credentials stay encrypted in Dataki, and every query runs with a read-only role.
- The data your app writes, not your code
- Dataki answers from the database: signups, orders, events. It never sees your repository.
More on the server itself, and every tool it offers, on the MCP server page. How credentials, keys and the query log work: security and governance.
Dataki and Claude Code
- How do I connect Claude Code to a PostgreSQL database?
- Connect the database to Dataki with a read-only user, create a Dataki API key, and run: claude mcp add --transport http dataki https://dataki.ai/api/mcp --header "Authorization: Bearer YOUR_DATAKI_KEY". Claude Code can then read the schema and run read-only queries through Dataki.
- Does Claude Code get my database password?
- No. Your database credentials are stored encrypted in Dataki and are never sent to Claude Code or to any model. Claude Code calls Dataki's MCP server, the server runs the SQL on its side with a read-only role, and only the rows that come back reach the conversation.
- Can Claude Code change or delete data through Dataki?
- No. Dataki's query tool is read-only: a statement that changes data is refused before it runs, and the database connection itself uses a role that cannot write. Claude Code can create and edit Dataki dashboards, which are Dataki's own records, never rows in your database.
- Which databases can Claude Code query through Dataki?
- Any source connected to Dataki: Google BigQuery, PostgreSQL, MySQL, Amazon Redshift, Supabase, Google Sheets and uploaded CSV or Excel files, including hosted databases such as Neon, Amazon RDS, Cloud SQL and PlanetScale. Google Analytics 4, Firebase, Search Console, Firestore and Stripe work through their exports to BigQuery. One question can read more than one source.
Connect Claude Code. Ask it something.
Connect a source to Dataki, add the server to Claude Code, and the first answer is one question away. Anything worth keeping becomes a dashboard with a link that stays live.