CData Arc 26.4 separates AI runtime actions from click-only diagnosis

CData Arc’s [October 27 release of version 26.4](https://arc.cdata.com/blog/cdata‑arc‑release‑q4‑2026) extends its Model Context Protocol server from inspection into operational actions. It is not Arc’s first MCP release: CData says the server arrived in the prior Q3 release with read‑only diagnostic tools. A compatible AI client can now request that Arc run a connector, requeue a failed message or clean up old files. Those operations change what the running system does, even though the release says they do not edit its configuration. The distinction matters for an operator deciding what to delegate: “no configuration change” is a narrower statement than “read‑only.” CData says each action requires explicit confirmation, stays within the user’s existing Arc permissions and is written to the audit log. These are the vendor’s descriptions of the controls; OpenTools has not tested their enforcement against a configured instance.

The same release adds a Diagnose error link to failed transactions in Activity, Message Detail and a connector’s Transactions tab. Clicking it asks the configured AI provider to analyze the failure and return an explanation with suggested corrective actions. CData says it runs only on that click and does not make changes. The result can be copied into a support case.

An explanation and an instruction to rerun work therefore belong to different paths. The diagnosis supplies suggestions; the MCP tools can perform the named runtime operations after the stated confirmation. A proposed correction should not be mistaken for a configuration fix that Arc has already applied.

The release also expands the evidence available for troubleshooting. Service‑level logs can expose failures, such as an SFTP Server connection failure, that happen before a transaction exists. For an existing message, Arc correlates application logs, request logs and transaction records. Looking only for a failed transaction can miss an earlier connection problem; correlating a message’s records addresses a different part of that investigation.

Analyze Worker Pool appears in Performance settings on the Advanced tab after an AI provider is configured. An administrator starts a logging session and selects its collection window. CData explicitly says collection starts from that point forward: it cannot capture the past, so the session needs to begin before the slowdown under investigation.

That timing is an important prerequisite, rather than something an AI explanation can repair afterward. Starting this particular collection session after a slowdown does not recreate the worker‑pool evidence from that earlier period. It can instead capture a later period in which the issue recurs.

After collection, Arc sends the session through analysis with the configured provider and returns findings and recommendations. CData says these remain suggestions, settings are not changed automatically and previous analyses are saved. Neither the worker feature nor transaction diagnosis should be described as an automatic performance‑tuning or repair system.

Both diagnostic workflows rely on the [configured AI provider described by CData](https://arc.cdata.com/blog/cdata‑arc‑release‑q4‑2026). The release does not establish that every analysis runs locally, nor provide complete data‑handling and cost terms for every possible provider. Installing Arc on premises should not be treated as proof of an all‑local AI‑processing arrangement. The availability instructions distinguish deployment types. On‑premises customers can download the current version from CData’s builds page; Arc Cloud customers are directed to support to schedule an upgrade. The announcement is not evidence that every cloud instance has already moved to 26.4. For operators, the useful sequence is to identify which tools can act, which only recommend, and whether the required logs were collected during the relevant period. The release adds those capabilities and prerequisites; it does not demonstrate a guaranteed root‑cause diagnosis, a measured speedup or independently verified permission isolation. Original illustrative figure by OpenTools Team, based on [CData’s October 27 Arc 26.4 release](https://arc.cdata.com/blog/cdata‑arc‑release‑q4‑2026). CData says MCP runtime actions require confirmation, use existing Arc permissions and enter the audit log; diagnostic tools instead return recommendations through the configured AI provider. Worker analysis requires logs collected prospectively. These are provider‑stated features, not independently tested controls. This is a simplified illustration, not product UI or an architecture diagram.