JetBrains just gave database agents a much better place to stand. DataGrip 2026.2 includes built-in MCP tools and skills for agent workflows, with JetBrains calling out Claude Code, Codex, and Junie support.
That is not the same as bolting a chat box onto an IDE. The useful bit is quieter: schema context, connection setup, text-to-SQL, and cleanup checks are becoming things an agent can request inside the database tool developers already use.
What DataGrip added
JetBrains lists five database tasks for the new agent workflow:
- Create data sources from a plain-text description, JDBC URL, or imported connection.
- Ask agents about schema structure in natural language.
- Generate SQL from natural-language requests using the live schema as context.
- Detect out-of-place tables and run dependency safety checks before cleanup.
- Mention specific database objects with
@dbObjector files with@fileName.
That is a sensible boundary. DataGrip is not promising that an agent should go run the database for you. It is giving the agent enough local context to help with ordinary database work: connect, inspect, query, refactor, document, and clean up carefully.
For teams already tracking MCP, this is the part to watch. The value is not another standalone assistant. The value is making the database environment legible to an agent without turning it loose.
Why built-in MCP tools matter
Database work is context-heavy. A generic model can write SQL, sure. But it does not know your schema, table relationships, naming conventions, indexes, stored procedures, or data source configuration unless someone feeds all of that into the prompt.
Nobody wants to paste schema dumps all day. It is slow, easy to get wrong, and usually missing the one relationship that matters.
MCP changes that workflow. The IDE can expose structured database context through tools. The agent can inspect the schema, look at objects, draft a query, and target the right file or database object with less guesswork.
The cleanup angle is the one I would treat most carefully. Schema cleanup gets dangerous fast if an agent cannot see dependencies. JetBrains highlighting dependency safety checks is a good sign, because database agents need guardrails before they get anywhere near destructive operations.
This also shows where MCP keeps spreading: not as a separate destination, but as a layer inside the tools people already keep open. Coding assistants, cloud consoles, browsers, data platforms, and now a database IDE.
What developers should take from it
Put the agent next to the system of record. A database agent that can inspect the actual DataGrip connection will usually beat a general assistant trying to reconstruct the schema from pasted notes.
Start with read-only work. Let the agent explore, draft SQL, explain tables, and write documentation. Cleanup tasks, migrations, deletes, and production writes should still require a human review step. If an agent can mention @dbObject, it can target a table precisely. That precision is useful. It also makes a bad instruction more expensive.
The other piece is composability. A DataGrip-style workflow can pair with directory skills such as SQL Generator, Schema Documenter, Data Cleaner, Database Migration Skill, and Accessibility Auditor for adjacent review work. The IDE provides live context. Skills provide repeatable procedures.
Database agents need boring controls
Database tooling is a hard test for agent adoption. Helpful agents can save hours on queries, schema drift, documentation, and migration planning. Careless agents can break production before lunch.
That is why this DataGrip release is worth paying attention to. It keeps the agent inside the developer workflow instead of floating above it. MCP exposes context, but the human still works in the IDE, where connections, objects, files, and review steps are visible.
Expect more database and data-platform products to copy this shape. The weak version is “AI writes SQL from a blank prompt.” The better version is simple: the agent sees the right schema, asks the right tool, drafts the change, and waits before doing anything risky.
FAQ
Q: Does DataGrip now support MCP? A: JetBrains says DataGrip 2026.2 includes built-in MCP tools and skills that give AI agents database capabilities inside the IDE.
Q: Which agents can use the DataGrip workflow? A: JetBrains specifically mentions Claude Code, Codex, and Junie.
Q: Is this only for text-to-SQL? A: No. Text-to-SQL is one use case, but the workflow also covers connection setup, schema inspection, database object mentions, and cleanup checks.
Q: Should agents be allowed to modify databases automatically? A: Not by default. Read-only inspection, query drafting, and documentation are good starting points. Cleanup, migrations, deletes, and production writes should require human review.