claude.ai connector
Add Basequel as a connector in claude.ai and approve access with OAuth. No keys to copy and paste.
Agent-native
Give your agents a real backend. Safely.
One connection reaches every Basequel project you choose. Tokens are scoped, SQL is read-only by default, and every call is logged.
Agent access is in active development. Early-access projects get it first.
Ways to connect
Bring the agent you already use.
Claude Code, Hermes and others
Any MCP-capable agent connects with a Basequel token. Create one per agent, scope it, revoke it when you’re done.
The basql CLI
The same capabilities from a terminal or a script. Useful for agents that prefer shell tools — and for you.
What agents can do
Work across every project, not one account.
An agent sees the projects its token allows — all of them, if you want — instead of whichever single account it signed in to.
- Find and list the projects it has access to
- Inspect schemas, tables and policies
- Query data with read-only SQL
- Make changes only when a token’s scope allows it
You
How many sign-ups did shop get this week compared with last week?
- describe_schema project=shopread
- run_sql read-onlyread
Agent
I counted new rows in
auth.users for both weeks and put the comparison in a table below.You
Great. Delete the test accounts while you’re there.
- run_sql writeneeds write scope
Agent
This token is read-only for shop, so I can’t delete anything. Here’s the query to run, or grant write scope and ask again.
Tool names are illustrative.
Safety
Guard rails by default. More access only when you grant it.
Agents are powerful and occasionally wrong. Basequel is built so a mistake stays small and visible.
Scoped tokens
Each token is limited to the projects and actions you choose. Give each agent its own and revoke it at any time.
Read-only SQL by default
Agents can look before they touch. Writes need a token that explicitly allows them.
Audit log
Every agent call is recorded: which agent, which tool, which project, under which token.
| Question | Answer |
|---|---|
| Which projects? | Only the projects the token lists. One project, several, or all of them. |
| Read or write? | Read-only unless you grant write access for that token. |
| Who did it? | Every call is in the audit log with the agent and token that made it. |
| How do I stop it? | Revoke the token or disconnect the connector. Access ends there. |
How it connects
Three steps, high level.
Authorise
Approve the claude.ai connector with OAuth, or create a token for your agent and choose its projects and scopes.
Call tools
The agent calls Basequel’s MCP tools or the
basqlCLI. Every call is checked against the token’s scope.Review
Every call lands in the audit log. Adjust scopes or revoke access whenever you like.
Exact connector URLs and install steps will be shared with early-access projects.
Early access
Be first to connect your agents.
Agent access reaches early-access projects first. Join the waitlist and tell us which agents you use.