v1.3.0

What's Changed

• Quick analysis now supports multiple images — \quick: true\ accepts an array (up to 5) for fast, cheap comparisons, not just single images. The single-image restriction and its error are removed; the quick default prompt was made plural-friendly.
• Tool descriptions made concise — the whole tool catalog (descriptions + schemas) is down to ~3,300 characters from ~8,500: redundant Args/Returns/Examples blocks, model IDs, and security details were removed since they duplicated the parameter schema. Titles shortened. Same decision-critical guidance, ~60% less context per injection.
• Description quality fixes — analyze_image now tells callers to state the comparison in the prompt when passing arrays; list_models gained explicit when-to-use guidance (finding valid model IDs when a configured model fails).
• Docs — README updated: \quick\ row and mode-selection table reflect multi-image quick analysis; fallback is now documented as detailed-mode only.

v1.2.0

What's Changed

• *Merged \�ision_helper_quick_analyze\ into \�ision_helper_analyze_image* — pass \quick: true\ for a fast, cheap single-image analysis (quick model, 1024-token output cap, minimal reasoning). One fewer tool in the catalog, no duplicated code.
• *Simplified \�ision_helper_list_models* — removed the
  esponse_format: json\ mode; markdown-only output with a leaner description.
• Retry fixes — \Retry-After\ is capped at 15 s, no wasted backoff sleep after the final attempt, and malformed 2xx responses fail immediately instead of being retried.
• Registry refresh — Windows registry values are re-read on a 60-second cycle, so an updated \setx OPENROUTER_API_KEY\ is picked up without restarting the client (matching documented behavior).
• Leaner tool descriptions — cut roughly 40-60% per tool since they duplicated the parameter schema docs.
• Tests + CI — 18 unit tests (registry parsing, MIME sniffing, image loading, error mapping, truncation) and a GitHub Actions workflow across Node 18/20/22.
• Docs — README updated for the merged tool and corrected working-directory/registry wording.

v1.1.1

Vision Helper MCP — stable release v1.1.1

What's new since 1.0.0

• vision_helper_quick_analyze — new tool for fast, cheap, low-latency image analysis. Defaults to meta/muse-glimmer-30b (configurable via OPENROUTER_QUICK_MODEL), caps output at 1024 tokens, and forces minimal reasoning effort.
• Automatic model fallback — vision_helper_analyze_image now retries with a fallback model (google/gemini-3.7-flash, configurable via OPENROUTER_FALLBACK_MODEL) when the primary model's provider is busy (429 / 5xx / timeout / network), so analyses don't fail.
• Truly detailed analysis — vision_helper_analyze_image now requests high reasoning effort and a higher default max_tokens (8192) so longer, thorough answers aren't cut short.
• Default model — changed to qwen/qwen3.8-max.
• Correct serverInfo.version — the MCP initialize response now reports the version from package.json instead of a hardcoded string.

Install / update

npm install -g vision-helper-mcp-server@1.1.1

The server is available on npm and GitHub at https://github.com/quickstraw/vision-helper-mcp-server.