What happened
DevoxxGenie creator Stephan Janssen announced version 1.15.0 on October 28, 2026. The release is available through the JetBrains Marketplace, where users can update it from Settings → Plugins. Janssen says most of the release was written by contributors outside the core team, and that two features were built by a Codex agent during the Devoxx Belgium 2026 opening keynote.
The model selector now supports favorites across providers. A user can star a selected model, then choose it from a Favorites entry; DevoxxGenie switches both provider and model. Availability is checked in the background. If a provider no longer offers a favorite, it remains listed but disabled. Favorites persist across IDE restarts and can be removed in the plugin's LLM Settings. Provider and model dropdowns are hidden when there is only one option. Separately, Scan & Copy settings that used to be global—including included extensions, excluded folders and files, .gitignore handling, and RAG directory exclusions—are now stored per project in its workspace file. A project without its own settings starts from a copy of the current global values. Scanning, copying a project, adding a project to a prompt, token counting, and RAG indexing use that project's settings.
The fetch_page tool now converts HTML to Markdown rather than returning visible text flattened into a single string. The conversion preserves headings, fenced code blocks with language labels, tables, lists, and absolute links; it removes navigation, footers, scripts, and hidden elements. When a page has a main element, only that part is used. The tool requests text/markdown first, so a site that serves Markdown directly can return it without HTML conversion. DevoxxGenie gives two examples: a LangChain4j tutorial grew from one line to 1,886 lines, while a Javadoc HttpClient page grew from 26 lines to 644.
Fetching is capped at 10 MB and 60 seconds. Redirects are followed only on the same host; a redirect elsewhere is returned as a URL for a separate tool call. The tool rejects URLs with credentials, hosts without a dot, and hosts resolving to loopback or link-local addresses, including localhost and 169.254.169.254. Pages are cached for 15 minutes. The release also adds Claude Fable 5.1, Opus 5.5, Opus 5, Sonnet 5, and Sonnet 5.5 to the Anthropic and AWS Bedrock provider lists, with pricing and context windows. Because these models reject non-default temperature, the plugin no longer sends that parameter to them.
The MCP update replaces the pinned langchain4j dependency with the newer client. Existing HTTP_SSE server configurations still load, but run over Streamable HTTP and are saved as HTTP the next time they are edited. The client negotiates protocol version 2026-07-28 when supported, otherwise 2025-11-25. Test Connection now uses the same transport as real execution. Servers that support only the old SSE protocol are no longer supported; users whose connections fail are advised to check for a Streamable HTTP endpoint. Other fixes include retaining the original prompt during long runs by evicting the oldest finished tool call, more resilient Ollama model loading, Bedrock Bearer Token authentication, and image delivery to CLI and ACP runners.
What the archive adds
We have not covered this before; it opens a new thread.
Trend signal
This release supports agent protocols become the interoperability layer: DevoxxGenie moves its MCP integration to Streamable HTTP and negotiates protocol versions rather than keeping an older transport pinned. The change is practical but bounded: it improves alignment with newer MCP clients while excluding servers that have only the legacy SSE endpoint.
Why it matters
For teams using DevoxxGenie, per-project Scan & Copy rules remove a source of accidental context leakage between repositories: a Java monorepo and a TypeScript project no longer have to share exclusions. The Markdown conversion makes fetched documentation more usable as agent input because code, tables, and links retain structure. Its size, time, redirect, and address checks also define concrete boundaries for a tool that retrieves arbitrary pages. Teams should still verify which parts of a page reach the agent and whether their documentation host is accessible under these limits.
The MCP transport change has an immediate migration implication. A configured server may continue to appear in settings and still fail if it exposes only the old SSE transport; teams should check for a Streamable HTTP endpoint before updating. Adding Claude 5 family models expands provider choices, but the release does not establish comparative quality or performance. The useful decision is operational: test the models and transport behavior in the actual projects and provider setup, rather than assume the new defaults fit every workflow.
What happens next
- Teams with legacy-only MCP servers will need to migrate or retain an earlier plugin version. (3 months, 70%). Watch for: Streamable HTTP endpoints or compatibility notices from configured server maintainers.
- Markdown retrieval will become the default input path for documentation-oriented agent tasks in the plugin. (3 months, 65%). Watch for: follow-on tool changes that expose extraction controls or retrieval results.
Limits and uncertainty
DevoxxGenie's announcement documents the feature behavior and gives examples of page-length changes, but it does not report independent tests of Markdown extraction quality across sites, retrieval accuracy, or the effect on agent task success. It does not quantify adoption or reliability of the new MCP transport. The announcement identifies servers that are incompatible when limited to legacy SSE, but does not provide a list of affected servers. Pricing and context windows are said to be included for the newly listed Claude models; the announcement does not reproduce those values, and they may vary by provider.
Skill to build now
If your team uses DevoxxGenie, inventory its configured MCP servers before updating: identify whether each offers Streamable HTTP, and test the connection with the plugin's updated Test Connection path. Then set project-specific Scan & Copy exclusions in a representative multi-project workspace and confirm that scanning, prompts, token counting, and RAG indexing use the intended rules. This turns the release's two most consequential workflow changes into a small, reversible compatibility check.
Reader takeaway
DevoxxGenie 1.15.0 makes context handling more project-aware and web retrieval more structured, but also changes the MCP compatibility boundary. JetBrains users should check their MCP servers for Streamable HTTP support and review per-project scanning rules before relying on the update.