[this benchmark] run 2 searches, Mon 2026-10-26, about 15:05 to 15:07 London (GMT), the WebSearch tool in standard mode (US
results). Run 1's five queries again. For each: the result URLs in order, and what the tool's summary said.
Pixelbolt or skulitom appeared in none. (freema/pixelforge-mcp on Glama is another project, not Pixelbolt's old name.)

1. "can Claude make pixel art sprites"
   shyft.ai/skills/pixel-plugin; cdn.jsdelivr.net willibrandon/pixel-plugin README; claudemarketplaces.com
   opusgamelabs/game-creator game-assets; spritecook.ai/blog/claude-desktop-spritecook-plugin (June 10, 2026);
   sourcepulse.org pixel-plugin; reddit.sentinel-team.org snapshot; spritecook.ai/docs/guide-generate-sprites;
   spritecook.ai/docs/ai-game-assets; claudepluginhub.com game-creator; claudemarketplaces.com/mcp/ai.spritecook/generate
   Summary: "Claude doesn't appear to draw pixel art natively"; it works "through tools, plugins, or generators".
   "I also found no independent quality comparisons." (Same ten URLs as 10-23, same order.)

2. "Claude Code pixel art MCP server benchmark"
   glama.ai/mcp/servers/mr9lhalsqc; glama.ai/mcp/servers/ly1xau145c; shyft.ai mcp-server-pixel-art; getdrio.com
   pixel-squads; claudemarket.ai/mcp/aitor-alt/pixel; designrevision.com best-mcp-servers-for-claude-code; glama
   steipete claude-code-mcp logo; getdrio.com mcp-server-pixel-art-kfischer-okarin; tomevault.io willibrandon/pixel-mcp;
   workingnotworking.com
   Summary: "I found no benchmark comparing pixel-art MCP servers for Claude Code." Named: the Glama "Pixel Art MCP"
   (Pollinations backend), pixel-mcp (npx, TypeScript), willibrandon's Aseprite pixel-mcp. The designrevision ranking
   is "opinionated ... not a lab benchmark" with no pixel-art metrics.

3. "does an MCP tool help Claude make better pixel art sprite animation"
   pkg.go.dev willibrandon/pixel-mcp; jsdelivr willibrandon/pixel-plugin README; glama.ai q5nfluslua, pkklvuhmfa;
   jsdelivr diivi/aseprite-mcp README; images.glama.ai freema/pixelforge-mcp; glama.ai j3sif2kleb, mr9lhalsqc;
   github.com/rkdfx/aseprite-mcp
   Summary: "Probably yes for workflow, but the results don't show it makes Claude's pixel art measurably better."
   "I found no independent benchmark comparing Claude with and without these tools on sprite animation quality."

4. "Claude Haiku Sonnet Opus pixel art sprite generation comparison"
   uxplanet.org (Haiku 5.5 vs Sonnet 5.5 vs Opus 5.5); docs.orq.ai changelog; morphllm.com; claw.aguidetocloud.com;
   haimaker.ai (twice); context.ai; canaltech.com.br; igmguru.com
   Summary: "No search result compares Claude Haiku, Sonnet, and Opus on pixel art or sprite generation." Its own
   inference: sprites would come from Claude writing code, "so a real comparison would need to test that workflow".

5. "AI agent pixel art sprite sheet GIF benchmark valid output palette"
   github.com/crsolver/claude-pxl/wiki; jsdelivr ccc007ccc/agent-paint-pixel-skills README; hackmd.io; emergentmind.com
   pixel-art-bench; shyft.ai pixel-plugin; wasmer mirror of willibrandon/pixel-mcp; pixie.haus (three pages);
   sorceress.games
   Summary: "No standardized benchmark for agent-made sprite sheets or GIFs with palette validity turned up." Closest:
   "Pixel Art Bench" (models answer a hex palette plus a character grid as JSON). Changed since 10-23: llamagen.ai's
   guide is gone; agent tools now rank (claude-pxl, agentpaint, Agent Sprite Forge for [another review tool]).

New names for Pixelbolt's next refresh (not read yet): crsolver/claude-pxl (.px.js source to PNG, sheets, GIF; an
agent skill), ccc007ccc/agent-paint-pixel-skills ("agentpaint", Rust CLI, APX/APXA formats, GIF export), Agent Sprite
Forge ([another review tool] skill with Python post-processing), Pixel Art Bench (emergentmind; a model benchmark, not a tool test).