How to Configure Windsurf Oracle Sandbox

Published
Length
3 min read

Luis Felipe Dias

All posts

You set up a windsurf oracle sandbox by creating a restricted database role first. Then you let the IDE agent run queries through your local shell or MCP server. Teams often start by giving the coding assistant their normal test database credentials. When those permissions are too broad, a single prompt can drop shared tables or read sensitive rows. How a windsurf oracle sandbox works in practice On June 2, 2026, Cognition renamed Windsurf to Devin Desktop with the same IDE features. Its agen

Your coding agents run with your access.

Claude Code reaches prod with a developer's credentials, and approval prompts get clicked through. hoop.dev checks every query, masks sensitive data, and holds risky changes for a human.

You set up a windsurf oracle sandbox by creating a restricted database role first. Then you let the IDE agent run queries through your local shell or MCP server. Teams often start by giving the coding assistant their normal test database credentials. When those permissions are too broad, a single prompt can drop shared tables or read sensitive rows.

How a windsurf oracle sandbox works in practice

On June 2, 2026, Cognition renamed Windsurf to Devin Desktop with the same IDE features. Its agent can run terminal commands on its own when the user gives permission. Those commands run inside a dedicated terminal that uses zsh on macOS or bash on Linux. Because the shell loads your profile, the agent inherits your local aliases and variables.

The legacy Cascade agent also talks to local MCP servers that you set up. Whether the agent uses the shell or MCP, it needs database credentials to run SQL. In a windsurf oracle sandbox, you create a dedicated user role inside the Oracle database. You grant that role select permissions on specific schemas and write rights only on scratch tables.

Engineers store the login details for that scoped Oracle role in their shell variables. When you ask the agent to test a schema change, it reads those variables. It then launches a command line tool or an MCP tool to run the query. The database checks the role permissions and returns the output back to the IDE.

How to scope database roles and shell permissions

Start by building a separate schema inside your non-production Oracle instance for agent tests. Create a new database user that only holds permissions for this single scratch schema. Do not grant admin roles or broad system rights to this test user account. If the agent writes a bad migration script, the blast radius stays inside that schema.

Next, review how the IDE agent runs commands on your developer machine. Keep user permission prompts turned on before the agent executes any terminal command. When the agent proposes a database command, read the SQL text before you approve it. You should also keep production credentials out of your default shell startup files.

If your shell loads production keys by default, the agent can read them during a session. Store only the sandbox credentials in the profile that the dedicated IDE terminal loads. This step stops the coding assistant from switching to a live system by mistake.

AI Sandbox Environments + End-to-End Encryption: architecture patterns and best practices

Continue reading? Get the full guide.

Free. No spam. Unsubscribe anytime.

What native sandbox setups solve and what they miss

Native database roles and IDE prompts solve the basic problem of who can log in. The Oracle role blocks the agent from touching schemas outside its assigned test area. The IDE prompt gives the developer a chance to stop obvious mistakes in the terminal. For small prototypes with fake data, these two native controls work well enough.

Still, test databases often hold copies of real customer emails, phone numbers, or card numbers. Once the Oracle role authenticates the session, the database trusts every allowed select query. Native role permissions do not rewrite sensitive columns in query results before returning them. When the agent runs a select statement, raw customer records flow straight into the prompt context.

Approving every terminal command by hand also wears teams down during long coding sessions. Developers start clicking approve without reading every line of generated SQL inside the prompt. Also, local shell logs do not give security teams a central view of agent queries. You know which database account logged in, but not which prompt triggered each statement.

Check database and API queries on the wire

Identity tools, roles, and network rules decide who opens a session. Setup alone cannot check what each query does or mask sensitive values in a response once a session starts. When workflows touch Postgres, MySQL, SQL Server, MongoDB, ClickHouse, Kubernetes APIs, SSH, or internal HTTP APIs, hoop.dev sits inline in the wire protocol beside those services as an open-source proxy, a sidecar for your Postgres or API. On every query an AI agent or engineer runs against production, hoop.dev decides to pass, mask, block, or hold it for human review.

Your login details and client tools stay as they are while hoop.dev guardrails refuse bad deletes or drops before they run, and hoop.dev data masking rewrites matched sensitive values in responses in memory. Furthermore, hoop.dev holds SOC 2 Type II, is MIT licensed, and writes every query and decision to an audit trail that generates evidence for HIPAA, GDPR, and PCI DSS. See the source on GitHub.

Frequently asked questions

Is Windsurf still separate from Devin Desktop?

No, Cognition announced on June 2, 2026, that Windsurf is now called Devin Desktop. It keeps the same IDE features, and the old docs links now point to Devin Desktop.

How does the IDE agent run Oracle queries?

The agent runs terminal commands with user permission in a dedicated zsh or bash shell. It can also use configured MCP servers that log in with an Oracle database role.

Why are static sandbox credentials risky for AI coding agents?

The dedicated terminal loads your shell profile and reads any keys stored there. If that role can read copied customer tables, raw emails and phone numbers enter the model context.

The query runs. The emails stay hidden.