Projects
Shape search engine
Find a Japanese kanji or Chinese hanzi by the parts you can see, with recursive search through decomposition trees, keyword and glyph lookup, English and Russian.
Glyph decomposition dataset
More than 24,000 characters (about 3,000 Japanese kanji and 21,000 Chinese hanzi, simplified and traditional) broken into named components, with alternate breakdowns, per-row history and documented provenance.
MCP server
A public, read-only Model Context Protocol server, so AI assistants can search characters by parts and fetch decompositions. No key needed.
Glyph rendering & audit tools
Command-line tools that render glyphs with headless Chromium, detect fonts silently substituting one glyph for another, and audit decompositions for phantom, missing or mislevelled parts.
Use it from code
The search API is public. For example, characters containing both "sun" and "moon":
curl -X POST https://kanji.alteon.help/kanji/api/search/parts \
-H 'Content-Type: application/json' \
-d '{"parts": ["sun", "moon"], "depth": 1}'Or add the MCP server to Claude Code:claude mcp add --transport http kanji https://kanjimcp.alteon.help/mcp
How we work
- Render it, don't reason about it. Every component is checked against the drawn glyph, not just against Unicode tables.
- Provenance first. Each piece of data records where it came from; open sources are credited and nothing is attributed to a source it did not come from.
- Tested. An isolated API test suite and more than 1,300 pinned decomposition checks run in CI on every push.