SShortSingh.
Back to feed

Atlassian MCP Server Permissions Versus Data Security Policies for AI Access

0
·1 views

A DEV Community author tested how to keep Claude's access to Jira and Confluence read-only through Atlassian's MCP server. Disabling the Jira write toolset in the Permissions tab blocked writes over both OAuth and API-token connections within 18 seconds, while reads continued. A data security policy blocking spaces or apps only filtered OAuth searches on the V2 endpoint and had no effect on the deprecated V1 endpoint. The author reports that the Permissions tab applies to the whole organization and was more reliable, whereas the data security policy requires Atlassian Guard, which costs $870 a month for 200 users. The post also notes that the settings have moved from Rovo to Connectors in the admin console, where they override Connected Apps settings.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingHacker News ·

Analogy Framing IPv4 Address Scarcity in Terms of Land

A blog post on iptrust.co presents IPv4 address scarcity through a land-ownership analogy. The article is shared on Hacker News, where it gained 6 points and has not yet drawn any comments. The source text provided contains only the title and link metadata, so the specific analogy details and conclusions cannot be verified from this item. The piece appears to explore how the finite supply of IPv4 addresses compares to a scarce, tradable resource such as land.

0
ProgrammingDEV Community ·

Batch API Costs Can Double: Six Pitfalls in a .NET OpenAI Batch Pipeline

A developer explains that OpenAI's Batch API offers half-price processing for non-urgent bulk work, but the surrounding application code can erase those savings. The article describes six common mistakes in a .NET 10 batch pipeline, including duplicate batch submissions after a crash. It recommends tagging each batch with metadata and saving state before every provider call so restarts don't resend work. It also advises sorting failures so only transient errors are retried, with a capped attempt budget, and notes that output lines can return out of order and may be missing. The piece is promotional, pointing readers to a paid course for the full working project.

0
ProgrammingDEV Community ·

Orquesta Lets Teams Collaborate on Environments via Agent Prompts, Without SSH

Orquesta was built to avoid the risks of sharing SSH keys, which expose entire filesystems and secrets, and which non-technical teammates cannot easily use. Instead, a local agent runs on the machine hosting the environment and executes prompts submitted from a web dashboard. The agent uses the Claude CLI, streams its output live, and commits every change to git, so no shell access is shared with users. Team members receive role-based permissions such as Admin, Developer, Reviewer, and Observer to control who can submit or only view work. The workflow lets a non-engineer request a change that the agent implements on a branch and submits as a pull request for review.

0
ProgrammingDEV Community ·

PostgreSQL JDBC driver could silently downgrade SCRAM channel binding, fixed in 42.7.12

CVE-2026-54291 affects the PostgreSQL JDBC driver when configured with channelBinding=require. Under that setting, the driver could fall back from SCRAM-SHA-256-PLUS to plain SCRAM-SHA-256 without informing the application. The connection could succeed while the client assumed the stronger protection, which is tied to the TLS tunnel, was in effect. The fix is included in JDBC driver version 42.7.12. Administrators are advised to upgrade all runtimes and bundled drivers and to verify the negotiated authentication mode through integration testing and logging.