PlanetScale's MCP server can now pull database branch metrics into an AI assistant

PlanetScale has added a tool to its Model Context Protocol (MCP) server that lets an AI assistant read performance figures for a database. The company's changelog entry, dated October 9, 2026, is titled "MCP server: Branch metrics" and says the server "can now check your database's metrics."

The tool is called planetscale_get_branch_metrics. According to PlanetScale, it pulls CPU, memory, disk, replication lag and connections into the assistant, so a developer can see those numbers without leaving their editor. Those five readings cover how hard a database server is working, how far a replica trails its primary, and how many clients are attached.

MCP is a common way for AI coding assistants to call outside tools. This entry adds a way for such an assistant to read operational health from PlanetScale. The practical change is that a question such as whether a branch is short on memory or lagging can be put to the assistant directly rather than answered in a separate dashboard. That is our reading of the announcement; the company's own wording is limited to the editor benefit quoted above.

Only one source supports this article: PlanetScale's own changelog. We did not find an independent report on the change, so everything here is the company's description of its own product, and we have not tested the tool.

Several details are not stated in the entry we reviewed. It does not say which database engines the tool covers, what time window the readings span, whether results are raw samples or summaries, or what permissions an assistant needs to call it. It also does not say whether organization settings that limit MCP query access, which PlanetScale announced the day before, apply to this tool. Readers who plan to rely on it should check PlanetScale's documentation for those points.

The update matters mainly to developers who already use PlanetScale from an AI-assisted editor. Seeing resource use and replication health next to the code being written could shorten the loop when diagnosing a slow branch. How well it works in practice, and how detailed the data is, remains unverified.

Revised 16 October 2026: Guth re-read the original sources and rewrote this article. The earlier version remains available as revision 1.